Chant SpeechKit
S2G is a universal software program that uses all features and functions of speech synthesizers. It records text acquired from any text file into audio file. If text is in graphic format S2G uses. A high quality speech synthesizer in various embodiments concatenates speech waveforms referenced by a large speech database. Speech quality is further improved by speech unit selection and concatenation smoothing.
Nuance is an American multinational computer software technology corporation, headquartered in Burlington, Massachusetts, on the outskirts of Boston, that provides speech recognition, and artificial intelligence. Nuance merged with its competitor in the commercial large-scale speech application business, ScanSoft, in October 2005. ScanSoft was a Xerox spin-off that was bought in 1999. The human voice, with all its subtlety and nuance, is proving to be an exceptionally difficult thing for computers to emulate. It will be a long, long time before a speech synthesis program.
You really don't have to sit in front of a computer with a mouse and keyboard to use information technology. Your applications can be enhanced to speak and listen to you from where ever you need them to.
Speech recognition is the process of converting an acoustic signal (i.e. audio data), captured by a microphone or a telephone, to a set of words. These words can be used for controlling computer functions, data entry, and application processing.
Speech synthesis is the process of converting words to phonetic and prosodic symbols and generating synthetic speech audio data. Synthesized speech can be used for answering questions, event notification, and reading documents aloud.
What is Speech Management?
Speech management enables you to:
- control application functions by speaking rather than having to use a mouse or keyboard,
- capture data by speaking rather than typing, and
- prompt and confirm data capture with spoken or audio acknowledgement.
Applications benefits include:
- enhanced speed and accuracy of data capture,
- added flexibility of running applications in a variety of environments, and
- expanded operating scenarios for hands-free computing.
What is SpeechKit?
Chant SpeechKit handles the complexities of speech recognition and speech synthesis to minimize the programming necessary to develop software that speaks and listens. Active dock 1 195.
It simplifies the process of managing Apple Speech, Google Android, Microsoft SAPI 5, Microsoft Speech Platform, Microsoft WindowsMedia (UWP and WinRT), and Nuance Dragon NaturallySpeaking recognizers, and managing Acapela, Apple AVFoundation, Cepstral, CereProc, Google Android, Microsoft SAPI 5, Microsoft Speech Platform, Microsoft WindowsMedia (UWP and WinRT), and Nuance Vocalizer synthesizers.
SpeechKit includes Android, C++, C++Builder, Delphi, iOS, Java, .NET Framework and Xamarin class libraries to support all your programming languages and sample projects for popular IDEs—such as the latest Visual Studio from Microsoft and RAD Studio from Embarcadero.
The class libraries can be integrated with 32-bit and 64-bit applications for Android and Windows platforms.
Speech Recognition and Synthesis Architecture
SpeechKit provides a productive way to develop software that listens. Applications set properties and invokes methods through a speech recognition management class. This class handles the low-level functions with speech recognition engines (i.e., recognizers).
Applications establish a session with a recognizer through which spoken language captured live via a microphone or from recorded audio is processed and converted to text. Applications use SpeechKit to manage the activities for speech recognition on their behalf. SpeechKit manages the resources and interacts directly with the speech application program interface (API). SpeechKit supports the following speech APIs for speech recognition:
- Apple Speech,
- Google android.speech,
- Microsoft SAPI 5,
- Microsoft Speech Platform,
- Microsoft System.Speech,
- Microsoft Microsoft.Speech,
- Microsoft WindowsMedia, and
- Nuance Dragon NaturallySpeaking.
Applications receive recognized speech as text and notification of other processing states through event callbacks.
SpeechKit Architecture for Speech Recognition
SpeechKit encapsulates all of the technologies necessary to make the process of recognizing speech simple and efficient for applications.
SpeechKit simplifies the process of recognizing speech by handling the low-level activities directly with a recognizer.
Instantiate SpeechKit to recognize speech within the application and destroy SpeechKit to release its resources when speech recognition is no longer needed.
Speech Synthesis Management
SpeechKit provides a productive way to develop software that speaks. Applications set properties and invoke methods through the speech synthesis management class. This class handles the low-level functions with text-to-speech engines (i.e., synthesizers or voices).
Applications establish a session with a synthesizer through which speech is synthesized from text. Applications use SpeechKit to manage the synthesizer resources on their behalf. SpeechKit manages the resources and interacts directly with a speech application program interface (API). SpeechKit supports the following speech APIs for speech synthesis:
- Acapela TTS,
- Apple AVFoundation,
- Cepstral Swift,
- CereProc CereVoice,
- Google android.speech.tts,
- Microsoft SAPI 5,
- Microsoft Speech Platform,
- Microsoft System.Speech,
- Microsoft Microsoft.Speech,
- Microsoft WindowsMedia, and
- Nuance Vocalizer.
Application receive notification of other processing states through event callbacks.
SpeechKit Architecture for Speech Synthesis
The ChantTTS class encapsulates all of the technologies necessary to make the process of synthesizing speech simple for your application. Optionally, it can save the session properties for your application to ensure they persist across application invocations.
SpeechKit simplifies the process of synthesizing speech by handling the low-level activities directly with a synthesizer.
Instantiate SpeechKit to synthesize speech within the application and destroy SpeechKit to release its resources when speech synthesis is no longer needed.
Feature Summary
Chant SpeechKit handles the complexities of speech recognition and speech synthesis. The classes minimize the programming efforts necessary to construct software that speaks and listens.
A SpeechKit application can:
- Control application functions by speaking rather than having to use a mouse or keyboard;
- Prompt users for applicable data capture;
- Capture data by speaking rather than typing;
- Confirm data capture with spoken or audio acknowledgement;
- Transcribe audio files to text; and
- Synthesize speech to files.
Recognizers are accessed via proprietary application programming interfaces (APIs). SpeechKit supports the following speech APIs for speech recognition:
| Speech API | Platforms |
|---|---|
| Apple Speech | ARM |
| Google android.speech | ARM |
| Microsoft SAPI 5 | x64, x86 |
| Microsoft Speech Platform | x64, x86 |
| Microsoft .NET System.Speech | x64, x86 |
| Microsoft .NET Microsoft.Speech | x64, x86 |
| Microsoft WindowsMedia (UWP) | ARM, x86, x64 |
| Microsoft WindowsMedia (WinRT) | x86, x64 |
| Nuance Dragon NaturallySpeaking | x86, x64 |
Synthesizers are accessed via proprietary application programming interfaces (APIs). SpeechKit supports the following speech APIs for speech synthesis:
Jfk Speech Synthesis
| Speech API | Platforms |
|---|---|
| Acapela TTS | x64, x86 |
| Apple AVFoundation | ARM |
| Cepstral Swift | x64, x86 |
| CereProc CereVoice | x64, x86 |
| Google android.speech.tts | ARM |
| Microsoft SAPI 5 | x64, x86 |
| Microsoft Speech Platform | x64, x86 |
| Microsoft .NET System.Speech | x64, x86 |
| Microsoft .NET Microsoft.Speech | x64, x86 |
| Microsoft WindowsMedia (UWP) | ARM, x86, x64 |
| Microsoft WindowsMedia (WinRT) | x86, x64 |
| Nuance Vocalizer | x64, x86 |
Within Chant Developer Workbench, you can:
- Enumerate speech engines for testing recognizer-, and synthesizer-specific features;
- Trace recognition and synthesis events;
- Support grammar activation and testing; and
- Support TTS markup playback.
SpeechKit License
You may explore the capabilities of Chant SpeechKit for 30 days. To continue to use the product after 30 days, you must purchase a license for the software or stop using the software and remove it from your system.
A valid purchased license gives you the right to construct executable applications that use the applicable class library and distribute it with executable applications without royalty obligations to Chant.
The Chant SpeechKit license is a single end-user license. Each developer who installs and uses SpeechKit to develop applications must have their own license.
SpeechKit class library names vary by platform: Windows 32-bit and 64-bit. This helps ensure the correct library is deployed with your application.
You may purchase a license for Chant SpeechKit on-line at the Chant store or through your preferred software reseller.
SpeechKit System Requirements
Development Environment
- Intel processor or equivalent,
- Microsoft Windows 10,
- 120 MB of hard drive space,
- CD-ROM drive,
- VGA or higher-resolution monitor,
- Apple Speech, Google android.speech, Microsoft SAPI 5 compatible, Microsoft Speech Platform, Microsoft WindowsMedia (UWP and WinRT), or Nuance Dragon NaturallySpeaking recognizer,
- Acapela, Apple AVFoundation, Cepstral, CereProc, Google android.speech.tts, Microsoft SAPI 5, Microsoft Speech Platform, Microsoft WindowsMedia (UWP and WinRT), or Nuance Vocalizer synthesizer,
- Android, C++, C++Builder, Delphi, iOS, Java, .NET Framework, or Xamarin development environment, and
- Close-talk microphone.
Where to Buy
Order Chant SpeechKit online at the Chant store or through your preferred software reseller.
Try ItBuy ItWelcome to the Nuance® Mix Platform. Nuance Mix enables you to build enterprise-grade multichannel solutions, with tooling and APIs that support your entire application life cycle and harness best-in-class conversational AI.
For changes between official releases, including monthly rollup announcements of new features, see the Change log.Last update: 2021-02-16
Feature highlights
The main features of this release are summarized below.
PII masking capabilities
Users of Mix.nlu and Mix.dialog can now mark entities and variables as sensitive to mask customer information in accordance with personally identifiable information (PII) standards. At runtime, sensitive data will be masked across all call logs for the application. By default, the sensitive attribute is 'off' (not sensitive) for all entities and variables. For details, see the documentation (change log), as well as the announcement on the Mix Community Forum.
Note: For changes to take effect, you need to build a new model and deploy to the applicable environments. Only then will the runtime services start masking the entities and/or variables.
User Service Credentials
Users can now create Service Credentials for their user profile. Similar to the Default credentials, these credentials are tied to the user's access levels across the various Mix resources such as projects, builds, and namespaces.
Service Credentials differ from Default credentials in that they make use of the Client Credentials OAuth 2.0 flow rather than the Authorization Code OAuth 2.0 flow. For more information, see Authorization in Mix. Hands off 3 0 5 download free. This feature simplifies service-to-service communication, as it removes the necessity of a manual login by instead relying on the client ID and client secret to provide a token.
Note: Currently, this feature is restricted to Global Admin and Global PS roles (in the Profile page, click Create Service Credentials). In future, this feature will be made available on an as-needed basis.
Mix.nlu Optimize tab
The Mix.nlu Optimize tab provides advanced automation tools to help power users more efficiently develop larger or more complex projects and to perform more sophisticated work on NLU models.
The Optimize tab (Sample Sentences panel) provides a unified view of all samples for the current language. You can filter samples by Intents, Entities, and Verification state, or by performing a Search using text or a regular expression. See the Mix.nlu documentation for details. Note that for users new to Mix.nlu, the Develop tab is still the best place to start developing models.
Message-level barge-in
Mix.dialog users can now control barge-in behavior on a per-message level. For example, disable barge-in to ensure that a legal disclaimer is played in its entirety without the possibility of interruption.
Barge-in is enabled by default, across all channels. The related setting has been removed from the global and channel levels in Project Settings, for simplicity. Users can now:
- Disable barge-in at the node level, for specific Question & Answer or Message nodes, to prevent the user from interrupting any of the messages in the node (configurable per channel); or
- Disable barge-in at the message level, in the Messages resource panel, to prevent the user from interrupting a specific message, wherever it is used.
Note: Message nodes do not yet let you disable barge-in per channel. When you disable barge-in in a message node, this applies to all channels.
See the 2020-12-08 Mix.dialog change log and the 2020-12-14 Dialog as a Service change log.
Mix.dialog graph performance improvement
The rendering of nodes and their transitions on the design canvas in Mix.dialog has been improved. By standardizing the height of nodes displaying messages, the graph is more easily rendered and thus more responsive, especially for larger projects.
Other new features and changes
This release introduces the following additional features and changes.
Languages and voices
New languages: Danish, Finnish, Mandarin (China), Mandarin (Taiwan), Swedish. For the complete list, see Languages.
New voices: For the Bengali, Bulgarian, Croatian, Czech, Dutch (Belgium), English (Great Britain, India, Ireland, Seychelles, South-Africa, United States), French (Belgium), Galician, German, Greek, Hebrew, Hungarian, Kannada, Polish, Portuguese (Brazil and Portugal), Romanian, Slovak, Swedish, Thai, Turkish, Ukrainian, and Valencian languages. For the complete list, see Text-to-Speech (TTS) voices.
Note: The Carmit voice incorrectly uses the language code 'he-IS' for Hebrew, Israel. In a future release, this will be corrected so that Carmit uses 'he-IL'. In the meantime, please use 'he-IS'.
Mix.dashboard
- Addition of project ID in filenames when exporting NLU and dialog models. For more information, see Export data.
- Performance improvements when importing NLU data (for example, samples in .txt format, models via TRSX).
- Improved error handling in the Add a Project dialog.
- Redesign of the Notifications console to improve the user experience.
- Ability to share project links with users who are members of the project, even if the users are not part of the same organization. See Share project links.
- Support for asynchronous application configuration status updates.
- Optimization of requests to the backend to improve performance.
Mix.nlu
- Ability to annotate entities when adding samples from the Try panel.
Mix.dialog
- Ability to convert intent components into generic components: Use this feature when you want to map multiple intents to the same intent component. In such cases, you must first convert the intent component into a generic component. Previously, the only way to achieve this was to export the JSON representation of your dialog model, edit it, and import it back into your Mix project. This can now be easily done within Mix.dialog with a single click. See Convert an intent component into a generic component.
- Ability to set locale-specific confidence thresholds: These tuning settings (documented in Global settings and behaviors) can now be set per language in multilingual projects:
- Collection settings: High- and low-confidence thresholds (at the project, channel, and entity levels)
- Confirmation settings: Low-confidence thresholds (at the project and channel levels)
- Ability to launch documentation at selected node: A new information icon in the upper-right corner of the Node properties pane opens the Mix documentation in a separate browser tab, with the relevant section for the node type in focus.
- Access to the last recognition literal for collection and confirmation: The predefined complex variables lastCollectionResultObject and lastConfirmationResultObject provide access to the collection literal directly, after each recognition. You might use these variables to reference the intent literal's length, such as in conditions. For example, to prompt users to provide a shorter answer if they are too verbose, or to send the full literal to an external service such as a search engine. See Manage variables for details.
- New predefined variables to hold the channel integration and user data: The predefined variables channelIntegration and userData have been added to all projects. The channelIntegration variable can be used to capture channel information provided by the client application, and the userData variable to collect data from the end user (time zone, location, user IDs, and so on) in a standard way to facilitate reporting. For example, in an IVR scenario userChannelID can be populated with a user's automatic number identification (ANI). See Predefined variables for more information. The client application sets these variables upon starting a session, as described in Exchanging session data.
Nuance runtime services
Proto files were updated for DLGaaS, NLUaaS, and ASRaaS. See the documentation change log for links, including to updated sections of the documentation.
Fixed issues
For Mix.dashboard:
- Under certain circumstances, ASR and NLU resources build successfully, however, dialog does not.
- Add a Project permits creation of a project with a custom channel that does not include a modality.
Known issues
No known issues identified since 3.5.0.
Feature highlights
The main features of this release are summarized below.
Mix.nlu Discover tab
The Mix.nlu Discover tab allows you to see what users are saying to your deployed application, giving you the opportunity to refine your NLU models based on actual data. Since the initial roll-out of the new Discover tab, continued enhancements have been made including the ability to:
- Sort the data by column such as Intent, Score, Collected On, and Region.
- Export (download) utterances to CSV (Comma Separated Value) format.
For more information, see Discover what your users say in the Mix.nlu documentation.
Mix.dialog intent mapping inheritance
The new intent mapping inheritance feature allows designers to globally configure intent mappings in the NLU resource panel. All intent mapper nodes in a project will dynamically inherit global mappings, but you have the option to set local mapping overrides at specific intent mapper nodes as needed. This new feature will be useful for projects that make use of multiple intent mapper nodes—for example, to route intents differently per line of business. For details, see the 9 October 2020 announcement on the Mix Community Forum, and Manage intents in the Mix.dialog documentation.
Other new features and changes
This release introduces the following additional features and changes.
Languages and voices
New languages: Korean, Malay, Vietnamese, and Indonesian. For the complete list, see Languages.
Additional new voices: New enhanced voices Isha-Mi (Indian English) and Kanya (Thai), and new standard voices including Mizuki (Japanese) and many more. For the complete list, see Text-to-Speech (TTS) voices.
Note that the following language codes have changed:
- Tiantian-Ml now uses cmn-CN
- Meijia-Ml now uses cmn-TW
- Sinji-Ml now uses yue-HK
Speaking style: Many Nuance voices now offer a style option, which sets an emotional tone for tagged sections of speech. Styles such as neutral, lively, forceful, and apologetic are available through the SSML style element. For example:
I'm sorry your parcel is late
Have a great day
For more information, see SSML tags in the TTS as a Service documentation.
Mix.nlu
- Rule-based entity type support: A rule-based entity defines a set of values based on a GRXML file. The ability to create a rule-based entity is only available to Professional Services (PS) and Global Administrator users. Although other users will see rule-based entities, they will not be able to modify the entity type nor the GRXML file associated with it.
- Additional improvements to the verification workflow.
Mix.dialog
General new features
- Access to the last interpretation result, including on nomatch: The literal of the last interpretation can now be accessed, including the list of recognized intents and their respective confidence scores. This variable, which is updated after each recognition, can be sent through a Data Access node for further processing and logic. See Predefined variables.
- Access to entity interpretation results: The 'Last Collection Interpretation' and 'Last Confirmation Interpretation' predefined variables have been expanded to return entity interpretation results, in addition to the intent interpretation results.
- New variable types to match NLU predefined entities: Alphanumeric (regex), Digits (regex), Decimal (nuance_DOUBLE), Amount (nuance_AMOUNT), Distance (nuance_DISTANCE), and Temperature (nuance_TEMPERATURE). These variable types have their respective methods and will enable additional formatting functionality in an upcoming release. See the Simple variable types topics in both the Mix.dialog and Dialog as a Service documentation.
- Access to confidence score of recognized intent: The confidence score of the active intent can now be accessed, in addition to its literal and value, providing more flexibility when handling intents and disambiguation in the tool. See Predefined variables.
- Display Send Data in Try mode: You can now see the data being sent to the data layer in Try mode, whether the data is sent through a Data Access node, External Actions node, or Question and Answer node.
- Anaphora resolution: Support in the latest release of the dialog context service for resolving anaphoras recognized by the NLU model, such as REF_TIME ('then') or REF_PLACE ('there'), into the corresponding, previously collected entity value (up to the two previous recognition steps).
- Confirmation messages: List entity and regex-based entity support for entity-specific confirmation messages.
IVR-related new features
- Grammar specifications: If your dialog includes nodes that are meant to collect entities whose values are determined by a grammar, you can export a grammar specification document that includes external grammar reference names, entity values, entity literals, corresponding DTMF keys, and so on. For details, see Manage grammars.
- Audio file IDs: In projects that include the Audio Script modality, it is possible to auto-generate audio file IDs based on the node and prompt type, for example, to reference audio files once they are deployed. See Manage messages for details.
- Dynamic message references: In projects that include the Audio Script modality, you can use the DynamicMessageReference predefined schema to insert a dynamic audio file reference that offers TTS as a backup. See the example in Predefined schemas.
- Event handling improvements: Ability to handle no text/audio (empty recognition) as a noinput event. In addition, a new default event
UserDisconnecthas been added so that hangup or user disconnect use cases can be handled as a standard event in the tool. - DTMF mapping: The DTMF tab of Question and Answer nodes allows you to create an entity on the fly, including values (comma-separated), which are automatically mapped to DTMF keys. See Set DTMF mappings. See below for support by DLGaaS.
Dialog as a Service
- DTMF mapping: DTMF to entity mappings are returned via the Question and Answer node in the ExecuteResponse. DTMF mappings can hence be consumed by a VoiceXML integration layer or application, as an alternative to referencing external grammars. See nuance.dlg.v1.common.DAAction.
- New variable types to match NLU predefined entities(Mentioned above for Mix.dialog): New variable types that match NLU predefined entities: Alphanumeric (regex), Digits (regex), Decimal (nuance_DOUBLE), Amount (nuance_AMOUNT), Distance (nuance_DISTANCE), and Temperature (nuance_TEMPERATURE). See Simple variable types.
New location for Mix documentation
Notice that the Mix documentation now has its own subdomain. Please bookmark this new location. Search and other functionality will continue to work seamlessly, and your previous bookmarks will be automatically redirected.
Deprecations and removals: TTS voices
The following deprecated language variants have been removed and are replaced by improved versions, as noted in Text-to-Speech voices.
| Removed language variant | Voice replacement |
|---|---|
| Zoe-Sc / Zoe-Mlsc | Zoe-Ml |
| Nathan-Ms | Nathan |
| Ava-Mls | Ava-Ml |
| Marisol | Marisol-Ml |
Fixed issues
- For Mix.dashboard:
- If you switch quickly from one namespace to another in the Organization area of the Manage tab, the list of users is not immediately refreshed.
- Redirect URLs entered in a user's Profile are not validated.
- Application deployment fails when the selected NLU and ASR builds were created at different times.
- For Mix.dialog:
- Emojis inserted in dynamic messages generate garbled text.
Note: Emojis must appear as Unicode escape sequences, in all messages. You can enter an escape string directly. If you insert an Emoji in its graphic form, it is now automatically converted to the escaped form. Emojis in existing messages are converted upon editing the message. - Message names: Some special characters, accented letters, and text in languages that don't use the Latin alphabet generate message names that cannot be modified.
- If a message node and a message have the same name, this generates an error.
- When you use the Entities tab of the NLU resource panel to add values for custom list entities, the new values are not available on the Interactivity tab of existing question and answer nodes that collect those entities.
- Emojis inserted in dynamic messages generate garbled text.
- For NLU as a Service:
- Potentially lower accuracy (and server error returned) for inputs that mix Japanese and non-Japanese characters.
Known issues
Known issues identified since 3.4.0 include:
- For Mix.dialog:
- Node-level recovery messages: Question and answer nodes do not yet support recovery messages for Max Turns events. When a MaxTurns event happens, the global Max Turns recovery message (or its channel-level equivalent, if any) is used.
- Currently, you can only map one intent to an intent component. In Mix.dialog version 3.5, if you wanted to map multiple intents to an intent component, you needed to first convert the intent component to a regular component: (1) Export the dialog model for your project, (2) in the exported JSON file, change
'description':'intent'to'description':'under the intent component that should receive multiple intent mappings, and then (3) import the updated JSON file into your project. This converted the intent component to a regular component, and allowed mapping multiple intents to it.
Note: Mix.dialog version 3.6 allows you to convert intent components into generic components via the Components pane (JSON workaround is not required). See Ability to convert intent components into generic components.
Feature highlights
Allmymusic 3 0 1 52. The main features of this release are summarized below.
Mix.dashboard
A new tab, called Manage, is now available in Mix.dashboard. This tab lets you perform all the tasks related to application management such as creating builds, creating and deploying application configurations, and managing organizations. The Manage tab replaces the Applications and Namespaces tabs. For more details about this feature, see the 2020-08-21 Change log.
You can create, and delete, multiple client IDs for the same app ID. See Create a client ID for details.
Mix.nlu
Mix.nlu now includes a verification workflow for project samples, giving you more control over which samples are used for your models. Samples automatically transition from the Intent verified to the Fully verified state when an annotation is added. This signals to Mix.nlu that you intend to add the sample to your model. You can always choose to assign a different state to the sample; for example, to exclude it (change the state to Excluded) or to use it to detect intent only (change to Intent verified). For more information, see Verify samples.
Note: Further enhancements are coming soon, including clearer verification state names, enhanced state indicators and easier state transitions, verification state memory for excluded samples, and verification state filtering.
Mix.dialog
Enhanced settings allow dialog designers to better control and augment the user experience. Use the new global confirmation settings to determine whether the app should prompt the user for confirmation when a recognized entity returns with a confidence score that is between the low and high confidence thresholds specified in the collection settings. You can reference the entity in focus during confirmation by inserting the dynamic Current Entity value placeholder in global confirmation messages. Predefined and custom entities now also have their own set of settings under each channel profile, so that the collection and confirmation behavior can further be tuned and tailored to the information being collected. Furthermore, it is now possible to override settings at the node level, in the Advanced tab of question and answer nodes, to further tune these settings locally and provide dialog designers as much control over the user experience as possible.
Note: With the current version of Mix.dialog, settings apply to all languages in multilingual projects.
Other new features and changes
This release introduces the following additional features and changes.
Languages and voices
New languages: English (India), Hindi (India), Italian (Italy), Portuguese (Portugal), Russia (Russia), and Spain (Spain). For the complete list, see Languages.
New voices: Hindi, Tamil, Telugu, Chilean Spanish, Columbian Spanish, Russian, Malaysian, and Vietnamese. For the complete list, see Text-to-Speech voices.
Mix.nlu
Support for locale-specific literal-value pairs in projects with more than one language/locale. For multilingual projects (projects with more than one language/locale), literal and value pairs are now locale specific. This means that if you add a value to a list-based entity, the data will be specific to the locale where this operation was performed. Similarly, if you delete a value this change will impact only the locale where the deletion was performed.
Mix.dialog
- Use an assign action to change the current intent value before returning to the intent mapper node. This allows switching intents manually—for example, after a set of disambiguation questions. You can also send the current intent value to an external system through data access nodes, external actions nodes, and question and answer nodes, if desired.
Note: Setting the current intent value manually in a component that has a question router node automatically triggers intent switching at the intent mapper node. However, if there is no question router node, the intent mapper node will use its On Return transition rather than navigate to the desired intent. Workaround: Add some logic to the On Return transition—for example, use a decision node and loop back to the intent mapper node when a Boolean variable (for example, isIntentSwitch) is true. - Support for regex-based entities. Use regex-based entities to collect sequences of digits or alphanumeric strings, such as zip codes or order numbers. Currently, some limitations have been observed when using regex-based entities for numbers that are triggered using audio.
- Ability to choose an Audio File Extension, in the Project Settings panel, for all channels or for specific channels that support the Audio Script modality. This file extension is appended to filenames in the audio file library to reference recorded messages at runtime through the VXML connector.
- Support for the DTMF modality. Ability to specify, for each channel profile, whether to allow referencing external speech or DTMF grammars in question and answer nodes. External grammars are referenced at runtime through the VXML connector. Speech grammars override NLU.
Note: You can set a grammar reference either as free text, or as a variable (for dynamic grammars). - Ability to override global settings and channel-specific settings, at the node level, in the Advanced tab of question and answer nodes, to further tune these settings locally.
Deprecations and removals
TTS as a Service: The v1beta1 protocol is obsolete and no longer supported.
TTS voices: The following language variants will be removed in an upcoming release, as noted in Text-to-Speech voices.
- For the Arabic voice Maged-MI, the ar-WW variant is scheduled for removal. Please use arb-001 instead.
- For the Norwegian voice Nora, the no-NO variant is scheduled for removal. Please use nb-NO instead.
Fixed issues
- For Mix.dashboard:
- The Projects list does not always sort as expected when filtered.
- When importing, it may take more time than expected for the entities to be updated in the drop-down presented in the Import Entity Literals dialog. When this happens, the list of entities may be empty or may reflect the list presented the last time the dialog was shown.
- Also when importing, the list of intents shown may not accurately present the possible choices.
- For Mix.dialog:
- If you import a multilingual project into a project that was created with a different set of languages, it is not possible to preview the dialog design.
Note: Mix no longer allows you to import a dialog model that specifies more languages than the project into which you want to import it. To be able to import a dialog model, the target project must have been created to support all languages specified in the JSON file. If you created the target project to support more languages, intentionally, edit the JSON file before importing it: add the codes for any missing languages to the 'supportedLocales' array, at the top level of the project JSON representation.
- If you import a multilingual project into a project that was created with a different set of languages, it is not possible to preview the dialog design.
Known issues
Known issues identified since 3.3.0 include:
- For Mix.dashboard:
- Under certain circumstances, ASR and NLU resources build successfully, however, dialog does not. (FIXED)
Workaround: Build dialog resources separately. - If you switch quickly from one namespace to another in the Organization area of the Manage tab, the list of users is not immediately refreshed. (FIXED)
- Under certain circumstances, ASR and NLU resources build successfully, however, dialog does not. (FIXED)
- For Mix.dialog:
- List entities and regex-based entities do not yet support entity-specific confirmation messages. For such entities, the global confirmation messages and any channel-specific overrides apply. (FIXED)
- Variables resource panel: Usage count for a variable doesn't reflect usage when the variable is used (1) in dynamic messages, or (2) as a return variable in external actions nodes (that is, you selected the variable under Get Data for a Transfer action). If your design doesn't reference such variables in any other contexts, you might end up deleting them by mistake.
- For NLU as a Service:
- Potentially lower accuracy (and server error returned) for inputs that mix Japanese and non-Japanese characters. (FIXED)
- Potentially lower accuracy (and server error returned) for inputs that mix Japanese and non-Japanese characters. (FIXED)
Feature highlight — Multi-lang projects
Nuance Mix now supports building a customer experience in multiple languages all within one Mix Project. At project creation, you can select which languages are supported and then activate/deactivate languages as needed as well as customize the experience inclusive of prompting and logic on a per-language basis. Finally, Mix.nlu supports adapting a centralized ontology with training data, annotations, and entity customizations on a per-language basis for optimized performance. For more specific details, see the 2020-06-11 Change log.
Other new features and changes
This release introduces the following additional new features and changes:
- The following new languages: Arabic, Dutch (Netherlands), and English (Australia). For the complete list, see Languages.
- For new voices, see the complete list of TTS voices available.
- Mix tooling support for right-to-left languages.
- For Mix.dashboard, ability to reset your Mix user password from the Profile page.
- For Mix.dialog:
- A nomatch is now thrown when the collected entity is not the entity in focus in the question and answer node or related to the current intent.
- Support for dynamic list entities (wordsets).
- Try tab bulk input of data access response data in JSON format.
- For all runtime services (XaaS), a new
user_idfield was added to all engines to provide greater granularity in logging and reporting. - For Dialog as a Service, support for custom events. See the Dialog as a Service gRPC API documentation change log for details.
Deprecated/obsolete protocols and endpoints
For all runtime services (XaaS) with the exception of Dialog as a Service (DLGaaS), the v1beta1 protocols will be removed in the next Mix release and v1beta2 (where applicable) will be removed in the following release. Please upgrade your applications to use the v1 protocols.
The beta endpoints will also be removed in the next Mix release. See URLs to runtime services for the production endpoints to use.
For DLGaaS, the v1beta1 and v1beta2 protocols have been removed. These endpoints are no longer available.
Fixed issues
- For Mix.dashboard:
- Under certain circumstances, the list of intents shown during an import does not correctly list possible choices.
- Sample credentials do not align with authorization mechanism.
- Unselecting a build when creating an application configuration does not disable the 'Set up Configuration' in all cases.
- Cannot export NLU model if project name contains Japanese characters.
- For Mix.dialog:
- If a message node and a message have the same name, this generates an error.
- Values added for custom list entities fail to appear on the Actions tab of existing question and answer nodes that collect these entities, which makes it impossible to set actions for the new values.
Known issues
Known issues identified since 3.2.0 include:
- For Mix.dialog:
- Mix.dialog version 3.2 (and earlier) allowed you to add messages for the Rich Text modality in channel profiles that don't support it. If you created a project that only supports TTS or Audio Script, and you didn't add specific message text for these modalities, you might be unable to see your existing messages in Mix.dialog. The design canvas shows Rich Text message text by default, if the current channel profile supports the Rich Text modality; otherwise TTS text appears if the channel profile supports it; otherwise Audio Script text appears.
Workaround: (1) Export the dialog model for your project, (2) in the exported JSON file, add 'Rich Text' to the 'modes' array, in the appropriate 'channel' object, under 'supportedChannels', and then (3) import the updated JSON file into your project. This will allow you to see the existing Rich Text version of your messages on the canvas and to use the Messages resource panel to copy and paste message text into TTS or Audio Script specific fields as needed. - For question and answer nodes, validation reports the warning 'Recognition node has no controller and defines no transition to another node in its default actions,' which you can safely ignore.
This warning may be encountered for question and answer nodes that collect NLU intents (and do not reference a question router node, which is a common situation), as well as for question and answer nodes that collect Boolean or yes/no entities, if there aren't any actions set in the Default section of the System Actions, and the node does not reference a question router node. - Question and answer nodes that collect isA relationship entities do not yet support value-specific messages and actions.
- When previewing a multilingual dialog design that allows switching to another language at runtime, messages that appear on the graph representation of the dialog flow in the Try tab remain in the language that was in effect when you started the preview session. (FIXED via documentation)
- If you imported a multilingual project into a project that was created with a different set of languages, it is not possible to preview the dialog design. Validation reports a warning (Query to bad url..), and an error (Error getting NLE model array..). (FIXED)
Workaround: Before importing a multilingual project into another, make sure the receiving project was created with the same languages as the originating project. - When you use the Entities tab of the NLU resource panel to add a value for a list entity, the new value is not available on the Interactivity tab of an existing question and answer node that collects that entity, which makes it impossible to define interactive elements for this value in this node. (FIXED)
Workaround: Add the new value directly from the question and answer node's User Input tab. If you already added several values from the Entities tab, it might be faster to reselect the entity being collected in the question and answer node. The latter is a destructive action that will remove any system actions and interactivity elements previously defined. Take note of these settings before reselecting the entity. - Assign actions: If you choose a variable or an entity, instead of typing an integer as the argument when using the
Integer.random(…)function in an assign action, the value returned by this function at runtime might be outside the expected range.
- Mix.dialog version 3.2 (and earlier) allowed you to add messages for the Rich Text modality in channel profiles that don't support it. If you created a project that only supports TTS or Audio Script, and you didn't add specific message text for these modalities, you might be unable to see your existing messages in Mix.dialog. The design canvas shows Rich Text message text by default, if the current channel profile supports the Rich Text modality; otherwise TTS text appears if the channel profile supports it; otherwise Audio Script text appears.
New features and changes
This release of Nuance Mix introduces the following new features and changes:
- Ability to exchange data with external systems. For details, see Exchange data with an external system.
- For new voices, see TTS voices. Note that the documentation now specifies all the languages supported by multilingual voices.
- For Mix.dashboard:
- Ability to import samples as Unassigned Samples.
- Nuance Data Pack (NDP) version now shown on the Details tab (together with QuickNLP/QNLP version).
- Link in UI to Contact Sales.
- Mix.nlu now permits upload of sample sentences to Unassigned Samples.
- For Mix.dialog:
- Ability to send data to the client via the question and answer node.
- Support for duplication of messages; for example, to conceptually 'unlink' a duplicated message from the original message.
- Support for node-level event handling.
- Support for custom events.
- Dialog as a Service now returns the ASR status and start-of-speech message.
- For ASR as a Service, note the following additions to the documentation:
- New section on resource weights.
- Examples of all audio formats.
- For TTS as a Service, a new field,
restricted, was added to the proto file to identify custom voices. - For Mix documentation, site-wide search is now available (in beta). Content reorganized to take a more task-based approach (many Mix.dashboard topics are now under Key Tasks).
Message-level barge-in
Mix.dialog users can now control barge-in behavior on a per-message level. For example, disable barge-in to ensure that a legal disclaimer is played in its entirety without the possibility of interruption.
Barge-in is enabled by default, across all channels. The related setting has been removed from the global and channel levels in Project Settings, for simplicity. Users can now:
- Disable barge-in at the node level, for specific Question & Answer or Message nodes, to prevent the user from interrupting any of the messages in the node (configurable per channel); or
- Disable barge-in at the message level, in the Messages resource panel, to prevent the user from interrupting a specific message, wherever it is used.
Note: Message nodes do not yet let you disable barge-in per channel. When you disable barge-in in a message node, this applies to all channels.
See the 2020-12-08 Mix.dialog change log and the 2020-12-14 Dialog as a Service change log.
Mix.dialog graph performance improvement
The rendering of nodes and their transitions on the design canvas in Mix.dialog has been improved. By standardizing the height of nodes displaying messages, the graph is more easily rendered and thus more responsive, especially for larger projects.
Other new features and changes
This release introduces the following additional features and changes.
Languages and voices
New languages: Danish, Finnish, Mandarin (China), Mandarin (Taiwan), Swedish. For the complete list, see Languages.
New voices: For the Bengali, Bulgarian, Croatian, Czech, Dutch (Belgium), English (Great Britain, India, Ireland, Seychelles, South-Africa, United States), French (Belgium), Galician, German, Greek, Hebrew, Hungarian, Kannada, Polish, Portuguese (Brazil and Portugal), Romanian, Slovak, Swedish, Thai, Turkish, Ukrainian, and Valencian languages. For the complete list, see Text-to-Speech (TTS) voices.
Note: The Carmit voice incorrectly uses the language code 'he-IS' for Hebrew, Israel. In a future release, this will be corrected so that Carmit uses 'he-IL'. In the meantime, please use 'he-IS'.
Mix.dashboard
- Addition of project ID in filenames when exporting NLU and dialog models. For more information, see Export data.
- Performance improvements when importing NLU data (for example, samples in .txt format, models via TRSX).
- Improved error handling in the Add a Project dialog.
- Redesign of the Notifications console to improve the user experience.
- Ability to share project links with users who are members of the project, even if the users are not part of the same organization. See Share project links.
- Support for asynchronous application configuration status updates.
- Optimization of requests to the backend to improve performance.
Mix.nlu
- Ability to annotate entities when adding samples from the Try panel.
Mix.dialog
- Ability to convert intent components into generic components: Use this feature when you want to map multiple intents to the same intent component. In such cases, you must first convert the intent component into a generic component. Previously, the only way to achieve this was to export the JSON representation of your dialog model, edit it, and import it back into your Mix project. This can now be easily done within Mix.dialog with a single click. See Convert an intent component into a generic component.
- Ability to set locale-specific confidence thresholds: These tuning settings (documented in Global settings and behaviors) can now be set per language in multilingual projects:
- Collection settings: High- and low-confidence thresholds (at the project, channel, and entity levels)
- Confirmation settings: Low-confidence thresholds (at the project and channel levels)
- Ability to launch documentation at selected node: A new information icon in the upper-right corner of the Node properties pane opens the Mix documentation in a separate browser tab, with the relevant section for the node type in focus.
- Access to the last recognition literal for collection and confirmation: The predefined complex variables lastCollectionResultObject and lastConfirmationResultObject provide access to the collection literal directly, after each recognition. You might use these variables to reference the intent literal's length, such as in conditions. For example, to prompt users to provide a shorter answer if they are too verbose, or to send the full literal to an external service such as a search engine. See Manage variables for details.
- New predefined variables to hold the channel integration and user data: The predefined variables channelIntegration and userData have been added to all projects. The channelIntegration variable can be used to capture channel information provided by the client application, and the userData variable to collect data from the end user (time zone, location, user IDs, and so on) in a standard way to facilitate reporting. For example, in an IVR scenario userChannelID can be populated with a user's automatic number identification (ANI). See Predefined variables for more information. The client application sets these variables upon starting a session, as described in Exchanging session data.
Nuance runtime services
Proto files were updated for DLGaaS, NLUaaS, and ASRaaS. See the documentation change log for links, including to updated sections of the documentation.
Fixed issues
For Mix.dashboard:
- Under certain circumstances, ASR and NLU resources build successfully, however, dialog does not.
- Add a Project permits creation of a project with a custom channel that does not include a modality.
Known issues
No known issues identified since 3.5.0.
Feature highlights
The main features of this release are summarized below.
Mix.nlu Discover tab
The Mix.nlu Discover tab allows you to see what users are saying to your deployed application, giving you the opportunity to refine your NLU models based on actual data. Since the initial roll-out of the new Discover tab, continued enhancements have been made including the ability to:
- Sort the data by column such as Intent, Score, Collected On, and Region.
- Export (download) utterances to CSV (Comma Separated Value) format.
For more information, see Discover what your users say in the Mix.nlu documentation.
Mix.dialog intent mapping inheritance
The new intent mapping inheritance feature allows designers to globally configure intent mappings in the NLU resource panel. All intent mapper nodes in a project will dynamically inherit global mappings, but you have the option to set local mapping overrides at specific intent mapper nodes as needed. This new feature will be useful for projects that make use of multiple intent mapper nodes—for example, to route intents differently per line of business. For details, see the 9 October 2020 announcement on the Mix Community Forum, and Manage intents in the Mix.dialog documentation.
Other new features and changes
This release introduces the following additional features and changes.
Languages and voices
New languages: Korean, Malay, Vietnamese, and Indonesian. For the complete list, see Languages.
Additional new voices: New enhanced voices Isha-Mi (Indian English) and Kanya (Thai), and new standard voices including Mizuki (Japanese) and many more. For the complete list, see Text-to-Speech (TTS) voices.
Note that the following language codes have changed:
- Tiantian-Ml now uses cmn-CN
- Meijia-Ml now uses cmn-TW
- Sinji-Ml now uses yue-HK
Speaking style: Many Nuance voices now offer a style option, which sets an emotional tone for tagged sections of speech. Styles such as neutral, lively, forceful, and apologetic are available through the SSML style element. For example:
I'm sorry your parcel is late
Have a great day
For more information, see SSML tags in the TTS as a Service documentation.
Mix.nlu
- Rule-based entity type support: A rule-based entity defines a set of values based on a GRXML file. The ability to create a rule-based entity is only available to Professional Services (PS) and Global Administrator users. Although other users will see rule-based entities, they will not be able to modify the entity type nor the GRXML file associated with it.
- Additional improvements to the verification workflow.
Mix.dialog
General new features
- Access to the last interpretation result, including on nomatch: The literal of the last interpretation can now be accessed, including the list of recognized intents and their respective confidence scores. This variable, which is updated after each recognition, can be sent through a Data Access node for further processing and logic. See Predefined variables.
- Access to entity interpretation results: The 'Last Collection Interpretation' and 'Last Confirmation Interpretation' predefined variables have been expanded to return entity interpretation results, in addition to the intent interpretation results.
- New variable types to match NLU predefined entities: Alphanumeric (regex), Digits (regex), Decimal (nuance_DOUBLE), Amount (nuance_AMOUNT), Distance (nuance_DISTANCE), and Temperature (nuance_TEMPERATURE). These variable types have their respective methods and will enable additional formatting functionality in an upcoming release. See the Simple variable types topics in both the Mix.dialog and Dialog as a Service documentation.
- Access to confidence score of recognized intent: The confidence score of the active intent can now be accessed, in addition to its literal and value, providing more flexibility when handling intents and disambiguation in the tool. See Predefined variables.
- Display Send Data in Try mode: You can now see the data being sent to the data layer in Try mode, whether the data is sent through a Data Access node, External Actions node, or Question and Answer node.
- Anaphora resolution: Support in the latest release of the dialog context service for resolving anaphoras recognized by the NLU model, such as REF_TIME ('then') or REF_PLACE ('there'), into the corresponding, previously collected entity value (up to the two previous recognition steps).
- Confirmation messages: List entity and regex-based entity support for entity-specific confirmation messages.
IVR-related new features
- Grammar specifications: If your dialog includes nodes that are meant to collect entities whose values are determined by a grammar, you can export a grammar specification document that includes external grammar reference names, entity values, entity literals, corresponding DTMF keys, and so on. For details, see Manage grammars.
- Audio file IDs: In projects that include the Audio Script modality, it is possible to auto-generate audio file IDs based on the node and prompt type, for example, to reference audio files once they are deployed. See Manage messages for details.
- Dynamic message references: In projects that include the Audio Script modality, you can use the DynamicMessageReference predefined schema to insert a dynamic audio file reference that offers TTS as a backup. See the example in Predefined schemas.
- Event handling improvements: Ability to handle no text/audio (empty recognition) as a noinput event. In addition, a new default event
UserDisconnecthas been added so that hangup or user disconnect use cases can be handled as a standard event in the tool. - DTMF mapping: The DTMF tab of Question and Answer nodes allows you to create an entity on the fly, including values (comma-separated), which are automatically mapped to DTMF keys. See Set DTMF mappings. See below for support by DLGaaS.
Dialog as a Service
- DTMF mapping: DTMF to entity mappings are returned via the Question and Answer node in the ExecuteResponse. DTMF mappings can hence be consumed by a VoiceXML integration layer or application, as an alternative to referencing external grammars. See nuance.dlg.v1.common.DAAction.
- New variable types to match NLU predefined entities(Mentioned above for Mix.dialog): New variable types that match NLU predefined entities: Alphanumeric (regex), Digits (regex), Decimal (nuance_DOUBLE), Amount (nuance_AMOUNT), Distance (nuance_DISTANCE), and Temperature (nuance_TEMPERATURE). See Simple variable types.
New location for Mix documentation
Notice that the Mix documentation now has its own subdomain. Please bookmark this new location. Search and other functionality will continue to work seamlessly, and your previous bookmarks will be automatically redirected.
Deprecations and removals: TTS voices
The following deprecated language variants have been removed and are replaced by improved versions, as noted in Text-to-Speech voices.
| Removed language variant | Voice replacement |
|---|---|
| Zoe-Sc / Zoe-Mlsc | Zoe-Ml |
| Nathan-Ms | Nathan |
| Ava-Mls | Ava-Ml |
| Marisol | Marisol-Ml |
Fixed issues
- For Mix.dashboard:
- If you switch quickly from one namespace to another in the Organization area of the Manage tab, the list of users is not immediately refreshed.
- Redirect URLs entered in a user's Profile are not validated.
- Application deployment fails when the selected NLU and ASR builds were created at different times.
- For Mix.dialog:
- Emojis inserted in dynamic messages generate garbled text.
Note: Emojis must appear as Unicode escape sequences, in all messages. You can enter an escape string directly. If you insert an Emoji in its graphic form, it is now automatically converted to the escaped form. Emojis in existing messages are converted upon editing the message. - Message names: Some special characters, accented letters, and text in languages that don't use the Latin alphabet generate message names that cannot be modified.
- If a message node and a message have the same name, this generates an error.
- When you use the Entities tab of the NLU resource panel to add values for custom list entities, the new values are not available on the Interactivity tab of existing question and answer nodes that collect those entities.
- Emojis inserted in dynamic messages generate garbled text.
- For NLU as a Service:
- Potentially lower accuracy (and server error returned) for inputs that mix Japanese and non-Japanese characters.
Known issues
Known issues identified since 3.4.0 include:
- For Mix.dialog:
- Node-level recovery messages: Question and answer nodes do not yet support recovery messages for Max Turns events. When a MaxTurns event happens, the global Max Turns recovery message (or its channel-level equivalent, if any) is used.
- Currently, you can only map one intent to an intent component. In Mix.dialog version 3.5, if you wanted to map multiple intents to an intent component, you needed to first convert the intent component to a regular component: (1) Export the dialog model for your project, (2) in the exported JSON file, change
'description':'intent'to'description':'under the intent component that should receive multiple intent mappings, and then (3) import the updated JSON file into your project. This converted the intent component to a regular component, and allowed mapping multiple intents to it.
Note: Mix.dialog version 3.6 allows you to convert intent components into generic components via the Components pane (JSON workaround is not required). See Ability to convert intent components into generic components.
Feature highlights
Allmymusic 3 0 1 52. The main features of this release are summarized below.
Mix.dashboard
A new tab, called Manage, is now available in Mix.dashboard. This tab lets you perform all the tasks related to application management such as creating builds, creating and deploying application configurations, and managing organizations. The Manage tab replaces the Applications and Namespaces tabs. For more details about this feature, see the 2020-08-21 Change log.
You can create, and delete, multiple client IDs for the same app ID. See Create a client ID for details.
Mix.nlu
Mix.nlu now includes a verification workflow for project samples, giving you more control over which samples are used for your models. Samples automatically transition from the Intent verified to the Fully verified state when an annotation is added. This signals to Mix.nlu that you intend to add the sample to your model. You can always choose to assign a different state to the sample; for example, to exclude it (change the state to Excluded) or to use it to detect intent only (change to Intent verified). For more information, see Verify samples.
Note: Further enhancements are coming soon, including clearer verification state names, enhanced state indicators and easier state transitions, verification state memory for excluded samples, and verification state filtering.
Mix.dialog
Enhanced settings allow dialog designers to better control and augment the user experience. Use the new global confirmation settings to determine whether the app should prompt the user for confirmation when a recognized entity returns with a confidence score that is between the low and high confidence thresholds specified in the collection settings. You can reference the entity in focus during confirmation by inserting the dynamic Current Entity value placeholder in global confirmation messages. Predefined and custom entities now also have their own set of settings under each channel profile, so that the collection and confirmation behavior can further be tuned and tailored to the information being collected. Furthermore, it is now possible to override settings at the node level, in the Advanced tab of question and answer nodes, to further tune these settings locally and provide dialog designers as much control over the user experience as possible.
Note: With the current version of Mix.dialog, settings apply to all languages in multilingual projects.
Other new features and changes
This release introduces the following additional features and changes.
Languages and voices
New languages: English (India), Hindi (India), Italian (Italy), Portuguese (Portugal), Russia (Russia), and Spain (Spain). For the complete list, see Languages.
New voices: Hindi, Tamil, Telugu, Chilean Spanish, Columbian Spanish, Russian, Malaysian, and Vietnamese. For the complete list, see Text-to-Speech voices.
Mix.nlu
Support for locale-specific literal-value pairs in projects with more than one language/locale. For multilingual projects (projects with more than one language/locale), literal and value pairs are now locale specific. This means that if you add a value to a list-based entity, the data will be specific to the locale where this operation was performed. Similarly, if you delete a value this change will impact only the locale where the deletion was performed.
Mix.dialog
- Use an assign action to change the current intent value before returning to the intent mapper node. This allows switching intents manually—for example, after a set of disambiguation questions. You can also send the current intent value to an external system through data access nodes, external actions nodes, and question and answer nodes, if desired.
Note: Setting the current intent value manually in a component that has a question router node automatically triggers intent switching at the intent mapper node. However, if there is no question router node, the intent mapper node will use its On Return transition rather than navigate to the desired intent. Workaround: Add some logic to the On Return transition—for example, use a decision node and loop back to the intent mapper node when a Boolean variable (for example, isIntentSwitch) is true. - Support for regex-based entities. Use regex-based entities to collect sequences of digits or alphanumeric strings, such as zip codes or order numbers. Currently, some limitations have been observed when using regex-based entities for numbers that are triggered using audio.
- Ability to choose an Audio File Extension, in the Project Settings panel, for all channels or for specific channels that support the Audio Script modality. This file extension is appended to filenames in the audio file library to reference recorded messages at runtime through the VXML connector.
- Support for the DTMF modality. Ability to specify, for each channel profile, whether to allow referencing external speech or DTMF grammars in question and answer nodes. External grammars are referenced at runtime through the VXML connector. Speech grammars override NLU.
Note: You can set a grammar reference either as free text, or as a variable (for dynamic grammars). - Ability to override global settings and channel-specific settings, at the node level, in the Advanced tab of question and answer nodes, to further tune these settings locally.
Deprecations and removals
TTS as a Service: The v1beta1 protocol is obsolete and no longer supported.
TTS voices: The following language variants will be removed in an upcoming release, as noted in Text-to-Speech voices.
- For the Arabic voice Maged-MI, the ar-WW variant is scheduled for removal. Please use arb-001 instead.
- For the Norwegian voice Nora, the no-NO variant is scheduled for removal. Please use nb-NO instead.
Fixed issues
- For Mix.dashboard:
- The Projects list does not always sort as expected when filtered.
- When importing, it may take more time than expected for the entities to be updated in the drop-down presented in the Import Entity Literals dialog. When this happens, the list of entities may be empty or may reflect the list presented the last time the dialog was shown.
- Also when importing, the list of intents shown may not accurately present the possible choices.
- For Mix.dialog:
- If you import a multilingual project into a project that was created with a different set of languages, it is not possible to preview the dialog design.
Note: Mix no longer allows you to import a dialog model that specifies more languages than the project into which you want to import it. To be able to import a dialog model, the target project must have been created to support all languages specified in the JSON file. If you created the target project to support more languages, intentionally, edit the JSON file before importing it: add the codes for any missing languages to the 'supportedLocales' array, at the top level of the project JSON representation.
- If you import a multilingual project into a project that was created with a different set of languages, it is not possible to preview the dialog design.
Known issues
Known issues identified since 3.3.0 include:
- For Mix.dashboard:
- Under certain circumstances, ASR and NLU resources build successfully, however, dialog does not. (FIXED)
Workaround: Build dialog resources separately. - If you switch quickly from one namespace to another in the Organization area of the Manage tab, the list of users is not immediately refreshed. (FIXED)
- Under certain circumstances, ASR and NLU resources build successfully, however, dialog does not. (FIXED)
- For Mix.dialog:
- List entities and regex-based entities do not yet support entity-specific confirmation messages. For such entities, the global confirmation messages and any channel-specific overrides apply. (FIXED)
- Variables resource panel: Usage count for a variable doesn't reflect usage when the variable is used (1) in dynamic messages, or (2) as a return variable in external actions nodes (that is, you selected the variable under Get Data for a Transfer action). If your design doesn't reference such variables in any other contexts, you might end up deleting them by mistake.
- For NLU as a Service:
- Potentially lower accuracy (and server error returned) for inputs that mix Japanese and non-Japanese characters. (FIXED)
- Potentially lower accuracy (and server error returned) for inputs that mix Japanese and non-Japanese characters. (FIXED)
Feature highlight — Multi-lang projects
Nuance Mix now supports building a customer experience in multiple languages all within one Mix Project. At project creation, you can select which languages are supported and then activate/deactivate languages as needed as well as customize the experience inclusive of prompting and logic on a per-language basis. Finally, Mix.nlu supports adapting a centralized ontology with training data, annotations, and entity customizations on a per-language basis for optimized performance. For more specific details, see the 2020-06-11 Change log.
Other new features and changes
This release introduces the following additional new features and changes:
- The following new languages: Arabic, Dutch (Netherlands), and English (Australia). For the complete list, see Languages.
- For new voices, see the complete list of TTS voices available.
- Mix tooling support for right-to-left languages.
- For Mix.dashboard, ability to reset your Mix user password from the Profile page.
- For Mix.dialog:
- A nomatch is now thrown when the collected entity is not the entity in focus in the question and answer node or related to the current intent.
- Support for dynamic list entities (wordsets).
- Try tab bulk input of data access response data in JSON format.
- For all runtime services (XaaS), a new
user_idfield was added to all engines to provide greater granularity in logging and reporting. - For Dialog as a Service, support for custom events. See the Dialog as a Service gRPC API documentation change log for details.
Deprecated/obsolete protocols and endpoints
For all runtime services (XaaS) with the exception of Dialog as a Service (DLGaaS), the v1beta1 protocols will be removed in the next Mix release and v1beta2 (where applicable) will be removed in the following release. Please upgrade your applications to use the v1 protocols.
The beta endpoints will also be removed in the next Mix release. See URLs to runtime services for the production endpoints to use.
For DLGaaS, the v1beta1 and v1beta2 protocols have been removed. These endpoints are no longer available.
Fixed issues
- For Mix.dashboard:
- Under certain circumstances, the list of intents shown during an import does not correctly list possible choices.
- Sample credentials do not align with authorization mechanism.
- Unselecting a build when creating an application configuration does not disable the 'Set up Configuration' in all cases.
- Cannot export NLU model if project name contains Japanese characters.
- For Mix.dialog:
- If a message node and a message have the same name, this generates an error.
- Values added for custom list entities fail to appear on the Actions tab of existing question and answer nodes that collect these entities, which makes it impossible to set actions for the new values.
Known issues
Known issues identified since 3.2.0 include:
- For Mix.dialog:
- Mix.dialog version 3.2 (and earlier) allowed you to add messages for the Rich Text modality in channel profiles that don't support it. If you created a project that only supports TTS or Audio Script, and you didn't add specific message text for these modalities, you might be unable to see your existing messages in Mix.dialog. The design canvas shows Rich Text message text by default, if the current channel profile supports the Rich Text modality; otherwise TTS text appears if the channel profile supports it; otherwise Audio Script text appears.
Workaround: (1) Export the dialog model for your project, (2) in the exported JSON file, add 'Rich Text' to the 'modes' array, in the appropriate 'channel' object, under 'supportedChannels', and then (3) import the updated JSON file into your project. This will allow you to see the existing Rich Text version of your messages on the canvas and to use the Messages resource panel to copy and paste message text into TTS or Audio Script specific fields as needed. - For question and answer nodes, validation reports the warning 'Recognition node has no controller and defines no transition to another node in its default actions,' which you can safely ignore.
This warning may be encountered for question and answer nodes that collect NLU intents (and do not reference a question router node, which is a common situation), as well as for question and answer nodes that collect Boolean or yes/no entities, if there aren't any actions set in the Default section of the System Actions, and the node does not reference a question router node. - Question and answer nodes that collect isA relationship entities do not yet support value-specific messages and actions.
- When previewing a multilingual dialog design that allows switching to another language at runtime, messages that appear on the graph representation of the dialog flow in the Try tab remain in the language that was in effect when you started the preview session. (FIXED via documentation)
- If you imported a multilingual project into a project that was created with a different set of languages, it is not possible to preview the dialog design. Validation reports a warning (Query to bad url..), and an error (Error getting NLE model array..). (FIXED)
Workaround: Before importing a multilingual project into another, make sure the receiving project was created with the same languages as the originating project. - When you use the Entities tab of the NLU resource panel to add a value for a list entity, the new value is not available on the Interactivity tab of an existing question and answer node that collects that entity, which makes it impossible to define interactive elements for this value in this node. (FIXED)
Workaround: Add the new value directly from the question and answer node's User Input tab. If you already added several values from the Entities tab, it might be faster to reselect the entity being collected in the question and answer node. The latter is a destructive action that will remove any system actions and interactivity elements previously defined. Take note of these settings before reselecting the entity. - Assign actions: If you choose a variable or an entity, instead of typing an integer as the argument when using the
Integer.random(…)function in an assign action, the value returned by this function at runtime might be outside the expected range.
- Mix.dialog version 3.2 (and earlier) allowed you to add messages for the Rich Text modality in channel profiles that don't support it. If you created a project that only supports TTS or Audio Script, and you didn't add specific message text for these modalities, you might be unable to see your existing messages in Mix.dialog. The design canvas shows Rich Text message text by default, if the current channel profile supports the Rich Text modality; otherwise TTS text appears if the channel profile supports it; otherwise Audio Script text appears.
New features and changes
This release of Nuance Mix introduces the following new features and changes:
- Ability to exchange data with external systems. For details, see Exchange data with an external system.
- For new voices, see TTS voices. Note that the documentation now specifies all the languages supported by multilingual voices.
- For Mix.dashboard:
- Ability to import samples as Unassigned Samples.
- Nuance Data Pack (NDP) version now shown on the Details tab (together with QuickNLP/QNLP version).
- Link in UI to Contact Sales.
- Mix.nlu now permits upload of sample sentences to Unassigned Samples.
- For Mix.dialog:
- Ability to send data to the client via the question and answer node.
- Support for duplication of messages; for example, to conceptually 'unlink' a duplicated message from the original message.
- Support for node-level event handling.
- Support for custom events.
- Dialog as a Service now returns the ASR status and start-of-speech message.
- For ASR as a Service, note the following additions to the documentation:
- New section on resource weights.
- Examples of all audio formats.
- For TTS as a Service, a new field,
restricted, was added to the proto file to identify custom voices. - For Mix documentation, site-wide search is now available (in beta). Content reorganized to take a more task-based approach (many Mix.dashboard topics are now under Key Tasks).
Fixed issues
- For Mix.dashboard:
- Data pack version information is not consistently displayed in the Details tab.
- Text entered in the search field to filter namespaces is not visible.
Known issues
Known issues identified since 3.1.0 include:
- For Mix.dashboard:
- Redirect URLs entered in a user's Profile are not validated. (FIXED)
- Projects list does not always sort as expected when filtered. (FIXED)
- When importing, under certain circumstances:
- It may take more time than expected for the entities to be updated in the drop-down presented in the Import Entity Literals dialog. When this happens, the list of entities may be empty or may reflect the list presented the last time the dialog was shown. (FIXED)
- List of intents shown does not accurately present the possible choices. (FIXED)
- Application deployment fails when the selected NLU and ASR builds were created at different times. (CANNOT REPRODUCE)
- For Mix.dialog:
- If a message node and a message have the same name, this generates an error. Validation reports: generated sentences conflict with prompt group names. (FIXED)
Workaround: Rename the message node or the message. - Project Settings: Enabling or disabling multiple settings in quick succession might affect your project's integrity.
Tip: Enable one setting, add the desired messages and actions for that setting, and then move on to the next setting. If you must disable multiple settings, allow 5-10 seconds between each. - Global command entity: Mix.dialog does not prevent you from diverting an entity that is already in use, and making it your global command entity, in Project Settings. Make sure the entity you choose to be the global command entity is not used for any other purposes.
- After you import the JSON file for a dialog model into a project, Mix.dialog shows the project name found in the JSON file, instead of the current project's name.
Workaround: Rename your project in Mix.dashboard. Adding one character and deleting it immediately is sufficient to restore the project name across all Mix tools. - Values added to Menu Type custom entities fail to appear on the Actions tab of existing question and answer nodes that collect these entities, which makes it impossible to set actions for the new values. (FIXED)
Workaround: Take note of any existing messages and actions in the Action tab, switch to the User Input tab, and choose the entity again. Finally, set all required messages and actions.
- If a message node and a message have the same name, this generates an error. Validation reports: generated sentences conflict with prompt group names. (FIXED)
New features and changes
This release of Nuance Mix introduces the following new features and changes:
- Portuguese language was added. See Languages for details.
- For new voices, see TTS voices.
- Access to the Mix community forum from all Mix authoring tools.
- For Mix.dashboard:
- Including Dialog in a build no longer requires also including an NLU build.
- Commit Version on Details tab has been temporarily removed.
- For Mix.nlu, predefined entities now reference the documentation for examples.
- For Mix.dialog:
- Ability to reuse existing messages.
- Enhanced message naming pattern based on the message contents makes them easier to reuse.
- Ability to add descriptions to nodes.
- Ability to nest conditions under If/Else if statements.
- Enhanced preview mode: When a conversation ends, a convenient button appears at the bottom of the chat pane allowing you to start a new session more easily.
- For Dialog as a Service:
- Ability to specify the voice to use for audio synthesis.
- Ability to provide confidence values for entities when sending interpretations from the client application.
- Escalation action, end action, and continue action now include an ID that identifies the node in the dialog application.
- For ASR as a Service, the proto files were renamed recognizer.proto, resource.proto, and result.proto, although the contents of these files remains the same.
- For NLU as a Service, the sample Java application was updated to the latest v1 protocol.
- For TTS as a Service:
- SSML elements are documented in Reference topics – SSML tags.
- The timbre code was added as an SSML element and native control code. See above link as well as Reference topics – Control codes.
Fixed issues
- For Mix.nlu: Moving a sample from an intent (including from NO_INTENT) to Unassigned Samples actually moves it to NO_INTENT.
- For NLU as a Service: Remove Unicode character u200e from all NLU gRPC interpret requests.
Known issues
Known issues identified since 3.0.0 include:
- For Mix.dashboard:
- Sample credentials do not align with the current authorization mechanism. (FIXED)
- Text entered in the search field to filter namespaces is not visible. (FIXED)
- Data pack version information is not consistently displayed in the Details tab. (FIXED)
- Unselecting a build when creating an application configuration does not disable the 'Set up Configuration' in all cases. Clicking the button in this scenario may hang the user interface. (FIXED)
- In some scenarios the Last Modified date on the Details tab shows the last used date, in other scenarios the project creation date.
- Importing the JSON file for a dialog model into a project does not update the Details tab and the Targets tab to reflect the channel profiles and languages from the source project.
Note: With the current version of Mix, importing the JSON file for a dialog model overwrites any channels and languages you chose when you created the target project. Mix.dialog correctly shows the channel profiles from the latest import but the languages remain those from when the project was created. Mix.dialog allows you to add messages in languages that your current design does not actually support. Attempting to use these languages in preview generates an error. If you want to add support for some or all languages that you had specified when you created the project, use this workaround: (1) Export the dialog model for you project; (2) in the exported JSON file, add the codes for the missing languages to the 'supportedLocales' array, at the top level of the project JSON representation; and then (3) import the updated JSON file into your project.
- For Mix.dialog:
- Users with viewer permissions cannot see conditions in a dialog design. They also cannot see advanced options for question router nodes.
- No mechanism to determine when a resource such as a custom event is in use, to avoid accidental deletion (for example, when an event handler relies on that event). (FIXED via documentation)
New features
This General Availability (GA) release of Nuance Mix introduces the following new features.
Mix product (overall)
- Mix 3.0.0 introduces a new version of the XaaS APIs, v1. The previous versions (v1beta1 and v1beta2) are still supported and your existing application configurations will continue to work, as long as you use the same application configurations. See the Runtime APIs Quick Reference.
- Mix 3.0.0 introduces new URLs to access Version 1 of the runtime services. Note that the Beta URLs are still available for the previous versions (v1beta1 and v1beta2). See URLs to runtime services for details (same page as above).
- URN format to load Mix resources has changed. See URN format (same page as above). The old format is supported but deprecated. This means that if you create a new application configuration, the new URN format will need to be used; however, your older configurations will continue to work as before.
- Nuance authorization service token expiration is now 15 minutes. For more information, see Access token lifetime.
- Introduction of Mix community forum at https://community.mix.nuance.com
Language support
The following new languages were added:
| Language | Language code |
|---|---|
| English (Great Britain) | en-GB |
| French (France) | fr-FR |
| Spanish (United States) | es-US |
For the complete list of languages, see Languages.
TTS voices
Many TTS voices were added. For the complete list of voices, see TTS voices.
Mix.dashboard
- Ability to create projects in one of the supported languages and to specify multiple channels and modalities.
- When creating a new project, you now have two options: to use a quick start with default characteristics or a custom setup.
- Introduction of Targets tab to view settings for locale, channels, and modalities.
- Performance improvements to project load times.
Mix.nlu
- Support for regex-based entities.
- Streamlined entities leveraged by Mix.dialog—reduced list to DATE, TIME, and YES_NO (in addition to predefined entities).
Mix.dialog
- Project permissions are applied—owner, admin, and viewer.
- Messages panel has an enhanced layout showing all messages and any channel-specific and format-specific variations as a table.
- Support for isA entity types. Improvements to handling of base entity types.
- Support for pseudorandom variable, for example, for random message selection to improve the conversational experience.
- Support for buttons (interactions).
- Support for new node type: The external actions node handles interactions with external systems outside of the dialog application. It supports actions to be performed when ending a conversation, transferring to another system, or escalating to a live agent, and allows exchanging information with external systems.
- Support for global settings per project/application. (Note: Confirmations are not currently supported.)
- Support for component-level event handling (such as no-match, no-input, help).
- Ability to control intent switching.
- Ability to access the current recognized intent value and literal (full user utterance) by using
currentIntent.value(for example 'checkBalance') andcurrentIntent.literal('I want to check the current balance of my savings account'). - Ability to check if a variable or entity is null—that is, it was cleared, or it has not yet received a value—to determine the next action in the dialog flow.
Nuance runtime services
New v1 protocols for all services. Earlier protocols are supported but deprecated, as noted earlier.
General
Rate limiting applies to all runtime services.
NLU as a Service (NLUaaS)
- Suppression of sensitive data (gRPC logging mode).
- Encryption of all models.
Dialog as a Service (DLGaaS)
- A new request and a new response are available, StreamInput and StreamOutput, to perform recognition on streamed audio using ASRaaS and provide speech synthesis using TTSaaS.
- New flag was added to the Start Request Payload to disable logging for ASR, NLU, TTS, and Dialog.
- The session timeout passed into the Start request is reset again after each execute request.
- Enhancements to handle language topics.
Documentation
AWS Polly
Three new additions to the documentation set:
- Get Started, which explains key concepts, provides a quick start to using Mix, and suggests high-level steps to consider when planning a client application.
- Data Packs, which provides detailed examples of the predefined entities included in the data pack versions supported by Mix (current version is 9.4.1).
- Runtime APIs Quick Reference, which summarizes the main resources for using the gRPC APIs.
For other documentation updates, see each document's change log.
Known issues
Below are known issues in this GA release.
Mix.dashboard
- Please do not use Japanese characters in your project name.
- If you sorted the project list, by project name (Z to A), by ID, or by last modification time, changing the namespace selection makes the project list unexpectedly revert to the default alphabetical order.
- Project versions currently apply to Mix.asr and Mix.nlu only. They do not apply to Mix.dialog for now.
- For projects that were previously created, you cannot modify the channel selections.
- Learn tab: Content is being revamped
- Building: Dialog build failures do not return specific warnings to users.
Nuance Speech Synthesis Program
Mix.nlu
- Entities prefaced with REF_ can potentially conflict with 'Referenced As' entities, also known as anaphoras. (Note: Anaphoras are not supported in Mix.dialog.)
- Mix.nlu does not support annotating a sample in a character-based language when it contains user-entered spaces. For example, you cannot annotate a Japanese sample if you typed spaces between the characters.
- Mix.nlu currently does not support annotating character-based languages with tag modifiers. For example, you cannot annotate Japanese samples with the AND, OR, and NOT modifiers.
- The console displays training status for the currently selected project only.
- Accuracy issues around duplicate samples in training data.
- Performance and latency sometimes stretched thin.
- Moving a sample from an intent (including from NO_INTENT) to Unassigned Samples actually moves it to NO_INTENT. (FIXED)
- Feedback loop: Discover tab is currently not operational. (FIXED)
Mix.dialog
- Variable assignments occur after Messages and before Transitions. In order to output a variable in a message, the assignment must be set in a previous node rather than in the node itself.
- If you rename an intent that is mapped to an intent component, although the mapping is preserved, the NLU resource panel lets you create a new intent component, which would break the existing mapping. Do not click Create Intent Component, for the renamed intent.
Note: It is momentarily not possible to rename intents in Mix.dialog. Avoid renaming intents in Mix.nlu.
Workaround: If you already clicked Create Intent Component by mistake, and still want to use your original intent component, converting the intent component into a generic component will let you restore the mapping between your renamed intent and the (now converted) component.
Nuance runtime services
NLU as a Service (NLUaaS)
- Remove Unicode character u200e from all NLU gRPC interpret requests. (FIXED)
A tokenize request with the Unicode character u200e will tokenize the request as if the character was not there; that is, it seems to ignore the character. Example: 'A u200e B' will be tokenized as 'A B'.
Documentation
Search is limited to the immediate page rather than to the entire documentation set. (FIXED)
The Mix Platform is now GA. See Release Notes for details.
Beta releases are detailed below:
Last update: 2020-04-03
Mix 3.0.0-beta4 — February 2020
This release of Mix introduces the following new features:
German (de-DE) language support.
ASRaaS and NLUaaS now support the IETF BCP 47 standard for specifying languages.
For Mix.nlu:
- UI change to Intents pane to apply a lighter theme.
- Dialog (predefined) entities transferred from the Custom Entities to the Predefined Entities section. Dialog predefined entities are read-only: they cannot be edited, renamed, or deleted.
In previous releases, you were permitted to edit, rename, and delete dialog entities (CC_EXP_DATE, CREDIT_CARD, CURRENCY, DATE, DIGITS, NATURAL_NUMBER, PHONE, SSN, TIME, YES_NO, and ZIP_CODE). Please note that if in a previous release of Mix:- You renamed a dialog entity, it will continue to appear in the Custom Entities list.
- You deleted a dialog entity, it will not appear in either the Custom Entities or the Predefined Entities list. It has been permanently removed.
- You edited a dialog entity to modify (or delete) its relationship, you will not be able to modify/undo your changes. Contact Nuance if this breaks your existing dialog flows.
For Mix.dialog:
- Support for mathematical expressions in variable assignments.
- Ability to set reentry messages in question and answer nodes. When the dialog flow returns to a node that has reentry messages, these messages play instead of the primary messages, which only play the first time.
- Support for methods in variable assignments: for example, .length(), .concat, .search, .substring, for strings; .getDay(), .getMonth(), .getYear(), .addDay, .addMonth, .addYear, for dates; .getHour(), .getMinute(), .addHour, .addMinute, for time values; .isBefore, .isAfter, and so on.
- Ability to include rich content, such as links and formatting code (HTML, SSML), in messages. Note that WYSIWYG rendering is not yet available.
- Support for nuance_BOOLEAN, nuance_CARDINAL_NUMBER, and nuance_ORDINAL_NUMBER builtins.
- Ability to use entity literals as values.
For ASRaaS: Speaker profiles are supported as URN references as part of speaker dependent acoustic model adaptation. See Speaker profiles.
For DLGaaS:
StartRequestandInterpretationeach contain a new field,session_timeoutandslot_literals, respectively. The proto files have been updated accordingly. See the DLGaaS documentation change log for more information.For TTSaaS: Support for raw Opus as well as Ogg-encapsulated Opus. A new proto file is available that includes new Opus fields. See the TTSaaS documentation change log for more information.
For other documentation updates, see each document's change log.
Known issues
- Mix.dialog and DLGaaS do not yet support anaphoras.
- Mix.dialog does not yet support isA entity types. (FIXED)
- In Mix.nlu, moving a sample from an intent (including from NO_INTENT) to Unassigned Samples actually moves it to NO_INTENT.
Mix 3.0.0-beta3 — January 2020
This release of Mix introduces the following new features:
- For Mix.dialog:
- Support for complex variables (schemas and data objects).
- Ability to rename components.
- Intent components appear as a separate list in the Components pane.
- Updates to supported browsers.
- For ASR as a Service:
- Smart routing: The Krypton ASR service, available from a separate URL, can handle requests in all languages supported by Mix.
- The raw Opus audio format is supported, along with Ogg-encapsulated Opus. See Opus audio format.
For documentation updates, see each document's change log.
Known issues
- For character-based languages such as Japanese, errors are no longer observed when annotating a sample with an isA or hasA entity. However, issues have been observed when:
- Annotating samples with the AND, OR, and NOT modifiers
- Annotating samples that contain spaces
- Using multilevel annotations (isA/hasA)
- An error occurs when attempting to delete an isA or hasA entity that refers to a renamed entity.
- The Mix.dialog tool currently does not provide the status of dialog-related jobs. (FIXED in 3.0.0-beta4)
- It is not currently possible to approve pending promotion requests. (FIXED in 3.0.0-beta4)
- In Mix.dialog:
- Complex objects can only be populated after data access calls (FIXED)
- The name of a data access node must not include spaces
Mix 3.0.0-beta2 — December 2019
This release of Mix introduces the following new features:
- Japanese (Japan) language support (in addition to eng-USA).
- New voices (along with Evan and Zoe, introduced in beta1). See TTS voices.
- For Mix.dialog, addition of entities as variables in both message and data access nodes; import/export robustness improvements; and ability to access a validation panel via the Tools menu (gear icon), documented in Validate your design. See also Mix.dialog improvements in the Change log (with highlights such as support for data access nodes, nomatch event handling, and ability to specify various states of fulfillment for an entity in node properties).
- Ogg Opus support (8kHz and 16kHz), in addition to PCM (8kHz and 16kHz), A-Law (8kHz), and Mu-Law (8kHz). Raw opus is not supported (ADDED in 3.0.0-beta3). For more information see the ASR as a Service gRPC API documentation.
- The documentation includes code samples in:
- ASR as a Service: Python and Go
- NLU as a Service: Python, Go, and Java
- Dialog as a Service: Python, Go, and Java
- TTS as a Service: Python only
- A new v1beta2 version of the Krypton protocol for the ASR as a Service gRPC API. (You may continue to use v1beta1 in your ASRaaS applications without any changes. However, you must use either the v1beta1 or v1beta2 protocol, not both.
- Various UI and robustness improvements.
Fixed issues
Text To Speech Synthesis
- In Mix.nlu, NOT operator is now functional.
- In Mix.dialog:
- AND/OR/NOT NLU operators are supported.
- Dialog projects can be renamed.
- Exported JSON can be imported successfully (without manual modification)
Known issues
- For character-based languages such as Japanese, an error occurs when annotating a sample with an isA entity, hasA entity, or tag modifier (AND, OR, NOT).
- If you've disabled cookies, you may get stuck at the Mix loading screen. To avoid this error, enable cookies in your browser.
Mix 3.0.0-beta1 — October 2019
For interim features introduced between beta1 and beta2, see the Change log.
Below are known issues in the 3.0.0-beta1 release.
Nuance Mix (product overall)
- Language support is limited to eng-USA, GEN (more added in 3.0.0-beta2)
- Renaming: Dialog projects do not get renamed (FIXED in 3.0.0-beta2)
- Cloning: Dialog projects cannot be successfully cloned (includes 'project versions')
- Building: Dialog build failures do not return warnings to users
- Application configuration: Overriding a configuration (with dialog) by pointing to a new NLU version will not use the correct model
- Promotion flow supports the Sandbox environment only
- Sample Apps: Currently only a Python sample app is provided (more added in 3.0.0-beta2)
- Ability to create sample projects is not fully functional
- Some links to documentation are incorrect (FIXED in 3.0.0-beta2)
- Learn tab content is being revamped
Mix.nlu
- Train/Build: The UI does not currently train or build multiple different sources (assumes all data, or one source, at train/build time) (FIXED in 3.0.0-beta2)
- Ontology and NLU: NOT operator is not functioning correctly (FIXED in 3.0.0-beta2)
- Feedback loop: Discover tab is currently not operational
Mix.dialog
This list not updated. See more recent features added.
- Import/Export
- Importing JSON overrides the project name
- Exported JSON requires changing 'data' top-level node to 'importProject' for successful import (FIXED in 3.0.0-beta2)
- All Nodes
- Entity-specific error and confirmation behaviors: No confidence-based confirmation
- Entity-specific validation is currently not supported
- For the Conditions panel, the ability to insert rows between existing conditions is limited. In addition, only one action per row/condition is supported at the moment.
- Entities
- Dynamic custom list and freeform concept (entity) types are not yet supported
- Basic functionality for predefined concepts is provided, with improvements coming in the future. Note that predefined concepts (entities) appear in the custom entities list.
- AND/OR/NOT NLU operators are not supported (FIXED in 3.0.0-beta2)
- Messages
- Currently text only is supported (audio, rich text, buttons, images, etc., coming in future)
- Re-entry prompt is not currently supported
- Randomization is not currently supported
- Variables
- Primitive data types are currently supported (complex/data objects coming in future)
- Data access
- Data access nodes are currently only supported via the Dialog as a Service (DLGaaS) API (FIXED in 3.0.0-beta2)
- Question Router Node
- No ability to supply message
- Multi-entity conditions and Skip Input are not fully supported
- Preview
- Validation of design is required for effective preview
- Logs and session data are currently not exposed
- Artificial data is currently not provided
Mix.api
The Mix.api, which gives programmatic access to Mix, is not yet available.
Nuance runtime services
There is currently no 'speech to meaning' entry point (ASR + NLU).
ASR as a Service (ASRaaS)
- Only eng-USA (GEN) is currently supported, as noted above (more added in 3.0.0-beta2)
- Codec support is limited to PCM (8kHz and 16kHz), A-Law (8kHz), and Mu-Law (8kHz). (Ogg Opus support, 8kHz and 16kHz, added in 3.0.0-beta2)
- No support for acoustic model adaptation per user (speaker profiles) (FIXED)
- No support for custom pronunciations
- No nightly data pack update/sync
NLU as a Service (NLUaaS)
- Only eng-USA language support, as noted above (more added in 3.0.0-beta2)
- Text-in supported (instead of Audio-in)
TTS as a Service (TTSaaS)
- Only eng-USA language support, as noted above (more added in 3.0.0-beta2)
- Available voices: Evan and Zoe (more added in 3.0.0-beta2)
Documentation
Search is limited to the immediate page rather than to the entire documentation set.
Mix 2.x to 3.0.0-beta1 change list
Below are some differences between Mix 2.x and Mix 3.0.0-beta1.
Nuance Mix (product overall)
Application configuration: App Key no longer exists. Instead, you use your App ID together with your Client ID to generate a client secret. Be sure to save your client secret somewhere safe, as you will not be able to access it again from Mix.dashboard.
Nuance Mix uses the OAuth 2.0 protocol for authentication
Mix.nlu
- New tokenization and normalization engine
- Latest NLU engine and research library
Mix.asr
- Latest ASR engine and research library
- 4000 OOV (Out Of Vocabulary) limit increased to a hard limit of 50000 OOV
Mix.dialog
- Entirely new dialog design tool, engine, and service
Nuance runtime services
- Support the gRPC protocol rather than WebSocket, with dedicated endpoints per service
- Require OAuth 2.0 token for authentication (NMAID/app key no longer needed)
- You use a URN to reference resources for services with the context tag; for example:
urn:nuance:mix/eng-USA//mix.asr: Domain language modelurn:nuance:mix/eng-USA//mix.nlu: NLU model
- Persistent upload via inline wordsets for ASR and NLU (replacing NDSP_CONCEPT_UPLOAD_*CMD)
ASR as a Service (ASRaaS)
- Dedicated asraas.beta.mix.nuance.com (instead of NCS command)
- Endpointing implicitly available
- Supports weighting of domain language models (vis-a-vis the underlying data pack)
- Supports inline wordsets (previously data upload)
NLU as a Service (NLUaaS)
- Dedicated nluaas.beta.mix.nuance.com (instead of NDSP_ASR_APP_CMD)
- Improved handling of text input
- Response includes mention confidence
- concepts renamed to entities
TTS as a Service (TTSaaS)
- Dedicated ttsaas.beta.mix.nuance.com (instead of NMDP_TTS_CMD/NVC_TTS_CMD/..)
Dialog as a Service (DLGaaS)
- Dedicated dlgaas.beta.mix.nuance.com (instead of NDSP_ASR_APP_CMD with flag)
- Various other differences
