Language-specific settings

To support language-specific settings and messages, choose the desired language from the menu near the name of the project.

Language-specific settings and messages are available in these categories:

• 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)
• Speech settings (ASR domain language model weight, at the project, channel, and node levels)
• TTS settings (model, gender, voice, at the project and channel levels)
• Collection default messages (at the project and channel levels), recovery messages (at the node level)
• Confirmation default messages (at the project, channel, and entity levels), confirmation messages (at the node level)

Configure the weight of the ASR domain language model

Self-hosted environments: This feature requires engine pack 2.2 (or later) for Speech Suite deployments; engine pack 3.10 (or later) for self-hosted Mix deployments.

This setting lets you tune the weight of the ASR domain language model built from samples defined in Mix.nlu, and referenced at runtime, when the Dialog service orchestrates with the ASR service. This is a language-specific speech setting, available at the project, channel, and node levels.

A value of 0.25 is appropriate for most purposes. In some use cases, increasing the weight to 0.50 might improve recognition accuracy for domain-specific vocabulary.

This is an advanced feature: setting the weight of the ASR domain language model very high or low can adversely impact ASR recognition accuracy.

1. Navigate to Speech settings, under All channels, or under the desired channel if the weight you want to configure is for a specific channel.
2. Expand the Domain Language Model weight list, and choose the appropriate value (between 0.05 and 1) for your application.
   

Configure TTS settings

You can configure global TTS settings, and channel-specific overrides, for each language in your project.

1. Navigate to TTS settings, under All channels, or under the desired channel if the settings you want to configure are for a specific channel.
   Initially, Public is selected, and the voices that are installed in your Mix environment for the active language appear.
   
2. Choose the desired voice.
   To narrow down the list if needed, choose the desired quality model (Enhanced, Neural, Standard), and the desired gender (Female, Male).
   If the voice you want to use doesn’t appear in this list but you know it will be available in the Mix environment where your client application is to be deployed, click Other, and enter the name, gender, and model, as they are defined for the desired voice.
   
   

   Alternatively, you can have the client application pass the desired voice, model, and gender, to the TTS service at runtime. See Voice, in the TTS as a Service gRPC API documentation, for details.

Configure collection default messages

You can define default messages to be used—globally, or for specific channels—when recognition events occur at the collection phase of question and answer nodes. See Define local recovery behaviors for more information on how these messages and other node-specific messages are used during a conversation.

You can add collection recovery messages for these events: No Match 1, No Match 2, No Match 3, No Input 1, No Input 2, No Input 3, Max No Input, Max No Match, Max Turns.

Use the View messages as selector to review any message variations by output modality, directly on the compact messages.

Examples of recovery messages

Recovery message	Description
No Match 1, No Match 2, No Match 3 pre-messages	To set some context before prompting the user again, in case of a nomatch event—for example Sorry, I didn’t get that. (with appropriate variations for the second and third nomatch events)
No Input 1, No Input 2, No Input 3 pre-messages	To set some context before prompting the user again, in case of a noinput event—for example Sorry, I couldn’t hear you. (with appropriate variations for the second and third noinput events)
Max No Input, Max No Match, Max Turns	You might, for example, notify the user that the conversation is about to end, due to the lack of usable input

Configure confirmation default messages

You can define default messages to be used when recognition events occur at the confirmation phase of question and answer nodes. See Define local confirmation behaviors for more information on how these messages and other node-specific messages are used during a conversation.

You can add global, channel-specific, and entity-specific confirmation messages for these events: Initial, No Match 1, No Match 2, No Match 3, No Input 1, No Input 2, No Input 3, Yes to confirm, No to confirm.

Use the View messages as selector to review any message variations by output modality, directly on the compact messages.

Examples of recovery messages

Confirmation message	Description
Initial	To prompt the user for a confirmation—for example, I understood X is that correct? where X is the entity being collected
No Match 1, No Match 2, No Match 3	To prompt the user again in case of a nomatch event—for example Sorry, I didn’t get that. Was it yes or no? (with appropriate variations for the second and third nomatch events)
No Input 1, No Input 2, No Input 3	To prompt the user again in case of a noinput event (with appropriate variations for the second and third noinput events)
Yes to confirm	To acknowledge a yes answer to the confirmation question—for example, Awesome, got it!
No to confirm	To apologize following a no answer to the confirmation question—for example, Sorry, my mistake…