Intelligent Media Services:AIAgentConfig

The language ID for ASR. Valid values:

• zh_mandarin: Chinese

• en: English

• zh_en: Chinese and English

• es: Spanish

• jp: Japanese

The silence detection threshold for sentence segmentation. A silence period longer than this duration triggers a sentence break. Unit: milliseconds. Valid range: 200 to 1200. Default: 400.

Controls the sensitivity of the voice activity detection (VAD) for interruptions. A higher value makes the agent harder to interrupt. Valid range: 0 to 11. Default: 11.

• 0: Disables VAD.

• 1-10: Adjusts the interruption sensitivity.

• 11: An enhanced mode with improved noise resistance and less impact on audio quality.

Specifies pass-through parameters for custom ASR integrations.

The minimum duration of voice activity, in milliseconds, required to trigger an interruption. This helps control interruption sensitivity. A value of 0 disables this feature. Valid range: 200 to 2000. A typical setting is between 200 and 500, which corresponds to 1 to 4 words. By default, this parameter is not set and the feature is inactive.

The ID of the voice to use for synthesis. Changes take effect on the next utterance. If not specified, the agent uses the default voice from its template. This parameter only applies to preset TTS voices. Maximum length: 64 characters. For available values, see Voice Demos.

Specifies the model ID. Currently, only minimax models are supported. Valid values: speech-01-turbo and speech-02-turbo.

Specifies the language ID. Currently, only minimax models are supported. By default, this parameter is empty. Setting this parameter enhances performance for the specified language or dialect. If you are unsure of the language, set the value to "auto" to enable automatic detection. Supported values include:

Specifies the emotion for the synthesized speech. Currently, only minimax models support this feature. Valid values:

• happy

• sad

• angry

• fearful

• disgusted

• surprised

• calm

The speech rate. Supported on all platforms.

The maximum number of conversational turns to retain in the LLM/MLLM history. Default: 10.

The system prompt for the LLM at the start of the call.

Parameters for Alibaba Cloud Model Studio (Bailian) applications, formatted as a JSON string. For parameter format details, see Alibaba Cloud Model Studio (Bailian) application parameters.

Additional query parameters for an OpenAI-compatible LLM. Parameters must be in key=value format, with multiple parameters separated by &. All values must be strings.

If enabled, the AI Agent sends the complete LLM result to the client after the full response is generated. This setting does not affect the streaming of subtitles.

The minimum length in characters for a text output chunk. Text shorter than this value is buffered. Valid range: 0 to 100. A value of 0 or an empty value (default) disables this limit.

The maximum delay in milliseconds before buffered text is forcibly sent. Valid range: 1000 to 10000. A value of 0 or an empty value (default) disables this limit.

Specifies whether the LLM message history should be synchronized with the content played by TTS. Default: false. When enabled, the saved history reflects the exact content played by the TTS, including interruptions.

[
  {"role": "user", "content": "Tell me a story."},
  {"role": "assistant", "content": "Sure, I can tell you a story from Romance of the Three Kingdoms. Would you<ims_agent_interrupted> like to hear it?"},
  {"role": "user", "content": "How about another one?"}
]

The model ID of the avatar.

Specifies whether to allow voice interrupt. Default: true.

The ASR processing policy when interruptions are disabled.

• cache: Caches the ASR text. The system processes the cached text in the next turn.

• discard: Discards the ASR text immediately.

By default, ASR text is cached.

Specifies whether to include the interruption keywords in the text sent to the LLM. Default: false.

Specifies whether to enable voiceprint recognition. Default: false. If you enable this feature, you must provide a valid voiceprint ID.

The unique ID for voiceprint recognition. By default, this is not set. You must register the provided voiceprint ID. For more information, see Register a voiceprint.

The mode for turn detection.

• Normal (Default): Does not use AI to determine semantic completeness.

• Semantic: Uses AI to determine if the user has finished speaking based on semantic context.

The pause detection time in AI mode. Unit: milliseconds. Default: -1.

• -1: The AI automatically determines an appropriate wait time.

• 0-10000: A custom wait time. We recommend a value between 0 and 1500 ms.

Controls how quickly the AI responds after detecting a pause. This parameter is only effective in Semantic mode.

• Low: Patient waiting. The AI waits up to 6 seconds, reducing the risk of interruption.

• Medium: Balanced waiting. The AI waits up to 4 seconds. Suitable for most scenarios.

• High: Quick response. The AI waits up to 2 seconds. This increases responsiveness but may also increase the risk of accidental cut-offs.

By default, this parameter is not set.

The ID of the ambient sound. You can obtain this ID from the advanced configuration section of the agent settings in the console.

The volume of the ambient sound. Valid range: 0 to 100. A value of 0 disables the sound.