stt_latency

STT service latency defaults.

This module contains P99 time-to-final-segment (TTFS) latency values for STT services. TTFS measures the time from when speech ends to when the final transcript is received.

These values are used by turn stop strategies to optimize timing. Each STT service publishes its latency via STTMetadataFrame at pipeline start.

All built-in values were measured with VADParams.stop_secs=0.2, the recommended default. If you change stop_secs, re-run the benchmark with your VAD settings and pass the measured value to your STT service constructor.

To measure latency for your specific deployment (region, network conditions, self-hosted instances), use the STT benchmark tool: https://github.com/pipecat-ai/stt-benchmark

Run the TTFS benchmark for your service and configuration, then pass the measured value to your STT service constructor:

stt = DeepgramSTTService(api_key=”…”, ttfs_p99_latency=0.45)