Conversational AI Metrics Callback
Tencent RTC Conversational AI provides rich metrics callback features. These status callbacks are sent through Tencent RTC's custom messages, allowing easy monitoring of conversation performance on the client side. Metrics such as LLM and TTS call duration and performance data are pushed to the terminal in real-time via Tencent RTC custom messages, enabling monitoring of Conversational AI duration and quality.
Through the Tencent RTC SDK Receive Custom Messages feature, you can listen for callbacks on the client to receive real AI dialogue metrics data. cmdID is fixed at 1.
Field | Type | Description |
type | Number | Message type, 10020 indicates AI service invocation callback |
sender | String | Sender's `userid`, which is the bot's ID |
payload | Object | Message payload, including metric details |
The payload object contains the following fields:
Field | Type | Description |
metric | String | The metric names are as follows: asr_latency llm_network_latency llm_first_token tts_network_latency tts_first_frame_latency tts_discontinuity interruption |
value | Number | Call metrics |
tag | Object | Tags associated with the metric |
The tag object contains the following fields:
Field | Type | Description |
roundid | String | Conversation round ID |
Metric Name Description:
Status Code | Description |
asr_latency | ASR latency. Note: The metric includes the time set by VadSilenceTime when starting an Conversational AI. |
llm_network_latency | Network latency of LLM requests |
llm_first_token | LLM first token latency, metric includes network latency |
tts_network_latency | Network latency of TTS requests |
tts_first_frame_latency | TTS first frame latency, metric includes network latency |
tts_discontinuity | Number of non-continuous TTS requests, indicating that after a TTS streaming request finishes playing, the next request has not yet returned a result, usually due to high TTS latency |
interruption | This conversation was interrupted |