STTMetrics are generated automatically by the LiveKit Agents framework and are emitted as events, not retrieved via direct API calls. To collect STTMetrics, set up an event listener on your STT instance: • Python Example:

from livekit.agents.metrics import STTMetrics

# Create your STT instance
stt = deepgram.STT()  # or another provider

def stt_metrics_wrapper(metrics: STTMetrics):
    asyncio.create_task(handle_stt_metrics(metrics))

stt.on("metrics_collected", stt_metrics_wrapper)

async def handle_stt_metrics(metrics: STTMetrics):
    print(f"STT Duration: {metrics.duration}")
    print(f"Audio Duration: {metrics.audio_duration}")
    print(f"Streamed: {metrics.streamed}")

• JavaScript Example:

stt.on(SpeechEventType.METRICS_COLLECTED, (metrics) => {
    console.log('STT Metrics:', {
        requestId: metrics.requestId,
        duration: metrics.duration,
        audioDuration: metrics.audioDuration,
        streamed: metrics.streamed,
        timestamp: metrics.timestamp
    });
});

STTMetrics properties include: • `audioDuration`: Duration of audio input (seconds) • `duration`: Time to create transcript (0 for streaming) • `streamed`: Boolean for streaming mode • `requestId`: Unique request identifier • `timestamp`: When metrics were recorded • `label`: STT provider identifier • `error`: Any error (optional) Metrics are emitted after each

recognize()

call for non-streaming STT, or on

RECOGNITION_USAGE

events for streaming STT. Listen for the

metrics_collected

event to capture and process STTMetrics in your application. Sources: livekit-agents/livekit/agents/metrics/base.py | metrics/send-metrics-to-3p/send_metrics_to_3p.py | agents/src/metrics/base.ts | STTMetrics | LiveKit Agents