Metrics

Metric	Meaning	Owner
session_create_ms	Time from session creation request to API response.	OpenTalking
asr_partial_latency_ms	Delay from user speech to first partial transcript.	OpenTalking + STT provider
llm_first_token_ms	Delay from text request to first LLM token.	OpenTalking + LLM endpoint
tts_first_pcm_ms	Delay from sentence submission to first PCM/audio bytes.	OpenTalking + TTS provider
avatar_first_frame_ms	Delay from audio submission to first avatar video frame.	OpenTalking + synthesis backend
render_fps	Video generation throughput of the synthesis backend.	synthesis backend
webrtc_first_frame_ms	Time until the browser receives the first playable video frame.	OpenTalking + WebRTC
av_drift_ms	Offset between audio and video playback timelines.	OpenTalking
queue_depth	Queue depth in the Worker or external model service.	OpenTalking / backend
steady_chunk_ms	Steady-state chunk inference time.	synthesis backend

Boundaries

• First-response metrics must state their start and end points.
• render_fps describes the synthesis backend only; it is not the full user-perceived FPS.
• Multi-card, NPU, or remote model-service runs must also record topology and queue state.