The audio path
An onboard MEMS microphone connects through the ES8311 audio codec and I²S. The shared board library owns this hardware setup. A useful first diagnostic reads a short capture and reports its signal level without saving or printing the audio.
The recorder and speech paths use 16 kHz mono signed 16-bit audio. Recording a message and recognizing a command are different jobs: the first collects samples; the second runs a model on them.
Practical interaction
Use a level meter so someone can tell whether the microphone hears them. Give clear waiting, speech, silence, stopped and submitting states. A no-speech timeout prevents a recording from hanging forever. A maximum duration bounds memory. VAD can decide when to stop after a pause, but a manual stop should remain available.
Wi-Fi bridge monitor behavior
After starting a message, the monitor allows ten seconds to begin speaking, ends after approximately five seconds of silence, and caps capture at thirty seconds. Silence-only captures are discarded. Recognition pauses during recording/submission. The recording is sent over Wi-Fi to the local bridge for transcription; this requires a configured service.
Physical meter sensitivity, actual thirty-second capture and running-agent message interaction still need device exercises. A displayed meter or successful build alone does not establish transcription quality.
USB Micro hold-to-talk and Voice Chat
Monitor 0.6.3 sends live USB microphone audio while Hold to talk is held, then silences it on release. Review the dictated prompt and press Send. Select the board’s TinyUSB UAC1 input in Desktop so these controls gate the microphone actually used for the conversation.
For continuous conversation, enable separate microphone keys in Desktop: first Push to talk, second Voice Chat, if offered. Then enable Hold to talk + Voice toggle in browser settings. Tap Voice to enable audio, tap again to silence it, or hold to request ending the chat. Board settings cannot apply or verify Desktop mappings. Voice Chat availability and second-key acceptance still need testing with the owner’s Desktop version.
Mic ON means live audio is flowing to a USB host stream. Mic OFF means outgoing USB silence; the sensor still samples locally. Waiting means the host has not opened its stream. Boot, reconnect and saving settings close audio. A muted board does not prove the desktop voice chat has ended.
Setup and control contract includes the tests, design choices and dated flash evidence.
Start here
Use audio-diag, then speech-diag or recorder-io-diag. locallink demonstrates LAN speech transcription; recorder keeps recordings on SD.
Brief your coding agent
First show a microphone-level diagnostic. Then add bounded recording with a manual stop, silence termination and visible feedback. Keep captured audio out of logs. Explain where audio goes and how failures appear.