Intelligent Media Services:AIAgentConfig

ASR language ID. Available options:

• zh_mandarin: Chinese

• en: English

• zh_en: Mixed Chinese and English

• es: Spanish

• jp: Japanese

Threshold for voice sentence segmentation detection. A segmentation is triggered when the silence duration exceeds this threshold. Valid range: 200ms to 1200ms. Default: 400ms.

Interruption threshold parameter. Valid range: [0, 11]. Default: 11.

• 0 means the VAD function is disabled.

• 1-10: a higher value makes interruption more difficult.

• 11 is significantly different from the previous values. The pre-processing causes less damage to conversation audio and provides stronger noise resistance.

Pass-through parameters for the in-house ASR integration.

Minimum duration threshold for voice activity detection, used to control interruption sensitivity. 0 disables this feature. Valid range: 200 to 2000 milliseconds. Commonly used values [200, 500] correspond to 1-4 characters. Default: empty, not applied.

Voice ID. Takes effect on the next sentence after modification. If not specified, the voice ID configured in the agent template is used. Only applies to built-in TTS. Input length must not exceed 64 characters. For available values, refer to Intelligent Voice Effect Examples.

Currently only minimax is supported. Available values: speech-01-turbo / speech-02-turbo

Currently only minimax is supported. Default: empty. Enhances recognition of specified minor languages and dialects. Once configured, it can improve voice performance in scenarios involving the specified minor language/dialect. If the minor language type is not clear, you can select "auto" to let the model determine the minor language type. The following values are supported:

Currently only minimax is supported. Minimax currently supports 7 emotions:

• happy

• sad

• angry

• fearful

• disgusted

• surprised

• calm: neutral

Supported on all platforms. For cosyvoice, the default is 1.0 with a valid range of 0.5-2.0. For minimax, the default is 1.0 with a valid range of 0.5-2.0.

Maximum number of LLM/MLLM conversation history rounds to retain. Default: 10.

The LLM system prompt used after the call is started.

Alibaba Cloud Bailian application center parameters, as a JSON string. For the parameter format, refer to: Alibaba Cloud Bailian Application Center Parameters

Additional query parameters for the OpenAI-protocol LLM. Parameters must use the key=value format. Multiple parameters are joined with &. All values must be strings.

When enabled, after the LLM produces a complete reply, the agent sends the complete LLM result to the client. This switch does not affect the streaming generation of subtitles.

Minimum length (in characters) of text output. Text shorter than this length is buffered for concatenation. Range: [0, 100]. 0 or empty means no limit. Default: empty.

Maximum delay time (in milliseconds) for text output. When this time is exceeded, the buffered text is forcibly output. Range: [1000, 10000]. 0 or empty means no limit. Default: empty.

Whether the large model message history should stay consistent with the TTS playback content. Default: false. When enabled, the saved large model messages stay consistent with the TTS playback content.

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

Model ID of the digital human.

Whether voice interruption is supported. Default: true.

ASR processing strategy under mode:

• cache: caches the ASR text. After the current turn ends, the cached ASR text is processed together in the next turn.

• discard: directly discards the ASR text.

The default behavior is to cache the ASR text.

Whether to keep interruption words and pass them to the LLM. By default, they are discarded.

Switch for whether to use voiceprint recognition. Default: false. When voiceprint recognition is enabled, a valid voiceprint ID must be passed in.

The unique identity ID for voiceprint recognition. Default: empty. The passed-in voiceprint ID must have been registered via the voiceprint registration interface. For the interface documentation, refer to: Register Human Voiceprint

Voiceprint registration mode. Default: Explicit

The mode for turn detection.

• Normal (default): standard mode. AI is not used to judge semantics.

• Semantic: uses AI to judge, based on contextual semantics, whether the speaker has finished speaking.

Pause judgment time under AI mode. Unit: milliseconds. Default: -1.

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

• 0-10000: custom wait time. It is recommended to set this between 0-1500 ms.

Only effective under Semantic mode. Controls how quickly the AI starts responding after a pause is detected:

• "Low": waits patiently, with a maximum wait time of 6 seconds, reducing the risk of being interrupted.

• "Medium": balanced wait (maximum wait time of 4 seconds), suitable for most scenarios.

• "High": fast response (maximum wait time of 2 seconds), improves speed but may increase the risk of incorrect cut-offs.

This field is empty by default.

Call ambient sound ID. Can be obtained from the advanced configuration of the agent settings in the console.

Volume of the call background sound. Valid values: [0, 100]. 0 means off.