melodic-vegetable-78787
07/28/2025, 5:57 AMtall-airline-16756
07/28/2025, 5:57 AMtall-airline-16756
07/28/2025, 5:58 AMDuration metric shows 0.0000s in Deepgram STT metrics, is expected behavior for streaming STT. Here's why:
• The Duration metric represents the time it took to create a transcript. For streaming STT, this is always set to 0.0 because the transcription process is continuous and real-time, rather than being tied to a single request.
• The Audio Duration metric, which increases (e.g., 5.0000s → 5.0500s), indicates the actual duration of audio being processed. This is the more meaningful metric for streaming scenarios.
• The Streamed flag being active confirms that the STT is operating in streaming mode.
However, the repeated metrics with the same Request ID and no transcription result suggest that while audio is being received, the transcription process is not completing. This could be due to:
• Network interruptions or latency affecting the streaming connection.
• Configuration issues with the Deepgram STT plugin.
• Potential bugs or limitations in the integration between LiveKit and Deepgram STT.
Since this issue does not occur with Cartesia but happens intermittently with Google STT and Deepgram STT, it may be specific to how these services handle streaming audio or their integration with LiveKit.
To investigate further, you may need to:
1. Check the network stability and ensure low latency for the streaming connection.
2. Review the configuration settings for the Deepgram STT plugin to ensure they match the recommended setup.
3. Examine the logs for any errors or warnings during the affected calls.
If the issue persists, it may require a deeper investigation into the integration or the STT service itself.
Sources: Logs, metrics, and telemetry | LiveKit Docs | livekit-agents/livekit/agents/metrics/usage_collector.py