Thanks — I’m responding to both parts: the minimum detectable voice volume, and why some users’ speech isn’t being picked up. Minimum detectable volume • There isn’t a single numeric dB minimum. Agents use Voice Activity Detection (Silero VAD) which decides “speech vs silence” from audio signal characteristics, not a fixed volume threshold. • VAD sensitivity is configurable (so effective detection level depends on those settings). Possible reasons user speech isn’t being detected • VAD sensitivity too low: the

activation_threshold

(default 0.5) may be set high so softer speech is ignored. Adjusting it increases/decreases required signal strength. • Short or fragmented speech:

min_speech_duration

and

min_silence_duration

thresholds can cause brief utterances to be missed. • Noise cancellation (BVC) or aggressive suppression: background/noise filtering can remove softer speech or interact poorly with certain headsets. • Device/SDK warnings or hardware issues: microphone drivers, built-in device suppression (earbuds), or SDK warnings about “silence detected on local audio track” can indicate capture problems even if audio eventually works. • Sample rate or format mismatch: VAD expects supported sample rates (8kHz or 16kHz); mismatches reduce detection reliability. • Missing pre-connect audio buffer: if users speak immediately on connect and

preConnectBuffer: true

isn’t enabled, early speech can be lost. • Network/device audio degradation: packet loss or degraded audio can reduce speech energy enough that VAD doesn’t trigger. What you can try (quick steps) 1. Lower the VAD

activation_threshold

or reduce

min_speech_duration

to make detection more sensitive. 2. Test with noise cancellation off to see if suppression is removing speech. 3. Ask affected users to try a different microphone or headset and confirm system microphone settings. 4. Ensure audio track is published correctly and look for any SDK “silence detected” warnings during publish. 5. If users speak immediately on join, enable

preConnectBuffer: true

so early audio isn’t lost. 6. Confirm sample rate is 8kHz or 16kHz for VAD compatibility and check for packet loss/network issues. If you’d like, tell me: • which VAD parameters you’re currently using (e.g.,

activation_threshold

,

min_speech_duration

), and • whether affected users are on specific devices or headsets I can recommend exact parameter values and next diagnostic steps. Sources: Agent speech and audio | LiveKit docs | Silero VAD plugin | LiveKit docs | SIP troubleshooting guide | LiveKit docs | Understanding "silence detected on local audio track" warning