AIAgentConfig

The language for ASR. Valid values:

• zh_mandarin: Chinese (Mandarin)

• en: English

• zh_en: Chinese-English mixed

• es: Spanish

• jp: Japanese

The maximum duration of silence in milliseconds before the ASR engine finalizes an utterance. A pause longer than this value signals a sentence break. Range: 200–1200. Default: 400.

The Voice Activity Detection (VAD) threshold for interruptions. Range: 0–11. Default: 11.

• 0: Disables VAD.

• 1–10: Sets the interruption sensitivity. A higher value makes the agent harder to interrupt.

• 11: An enhanced mode with lower audio distortion and stronger noise resistance.

Passthrough parameters for proprietary ASR integrations.

The minimum duration in milliseconds of continuous user speech required to trigger an interruption. This controls interruption sensitivity. A value of 0 disables this feature. Range: 200–2000. A common range is 200–500 ms, which typically corresponds to 1 to 4 Chinese characters. If omitted, this feature is disabled.

The ID of the preset TTS voice. Changes apply to the next utterance. If omitted, the voice from the AI agent template is used. The ID can be a maximum of 64 characters. For available voices, see Intelligent Voice Samples.

This parameter applies only to the Minimax provider. Valid values: speech-01-turbo, speech-02-turbo

This parameter is for the minimax provider only. It enhances recognition for specific low-resource languages and dialects. If the language is unknown, set this to auto for automatic detection. By default, this parameter is not set. Supported values include:

This parameter applies only to the Minimax provider. Supported emotions include:

• happy

• sad

• angry

• fearful

• disgusted

• surprised

• calm

The speech rate, where a value of 1.0 is normal speed. The supported range can vary by provider. For CosyVoice, the range is 0.5 to 2.0 (default: 1.0). For Minimax, the range is 0.5 to 2.0 (default: 1.0).

The maximum number of recent conversational turns to include in the LLM/MLLM context. Default: 10.

The system prompt for the LLM after the call starts.

Parameters for Alibaba Cloud Model Studio, provided as a JSON string. For the parameter format, see Alibaba Cloud Model Studio Parameters

Additional query parameters for an OpenAI-compatible LLM. Parameters must be provided as a URL query string (e.g., key1=value1&key2=value2). All values must be strings.

When set to true, the agent sends the entire LLM response in a single message after it is fully generated, rather than streaming it. This setting does not affect the streaming of subtitles.

The minimum number of characters in a text chunk before it is sent to the TTS engine. Shorter chunks are buffered. Range: 0–100. A value of 0 or omitting this parameter disables buffering. Default: Not set.

The maximum delay in milliseconds before buffered text is sent to the TTS engine, even if OutputMinLength is not met. Range: 1000–10000. A value of 0 or omitting this parameter disables the delay limit. Default: Not set.

Specifies whether the LLM message history is synchronized with the content played by the TTS. Default: false. When enabled, the saved LLM messages match the content actually played by the TTS.

[
  {"role": "user", "content": "Tell me a story."},
  {"role": "assistant", "content": "Okay, I can tell you a story about the Three Kingdoms. Would you<ims_agent_interrupted> like that?"},
  {"role": "user", "content": "Tell me a different one."}
]

The model ID of the avatar.

Specifies whether to enable speech interruption. Default: true.

Specifies how to handle user speech that occurs during a non-interruptible section of the agent's utterance.

• cache: Caches the user's speech and processes it in the next conversational turn.

• discard: Discards the user's speech.

Default: cache.

Specifies whether to include the interrupt words in the text sent to the LLM. Default: false (words are discarded).

Specifies whether to enable voiceprint recognition. Default: false. If set to true, you must also provide a valid VoiceprintId.

The unique identifier for the voiceprint. This is not set by default. The ID must correspond to a voiceprint registered using the voiceprint registration API. For more information, see Register a voiceprint.

The voiceprint registration mode. Default: Explicit.

The conversational turn detection mode.

• Normal (Default): The agent relies on pauses to detect the end of a user's turn.

• Semantic: The agent uses AI to analyze conversational context to determine if the user has finished speaking.

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

• -1: The AI automatically determines a suitable wait time.

• 0–10000: A custom wait time. A range of 0–1500 ms is recommended.

Controls the agent's response speed after detecting a user pause. This parameter applies only in Semantic mode. A higher setting results in a faster response but increases the risk of interrupting the user:

• Low: Waits patiently with a maximum wait time of 6 seconds, reducing the risk of interruption.

• Medium: A balanced wait time (up to 4 seconds), suitable for most scenarios.

• High: Responds quickly (up to 2 seconds), which improves speed but may increase the risk of incorrect turn-taking.

This field is empty by default.

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

The volume of the ambient sound. Range: 0–100. A value of 0 disables the sound.