Two-path microphone comparison

No bot playback or conversation. Audio is saved locally for this diagnostic only. Use the same explicitly selected microphone for both recordings. Maximum 30 seconds per recording.

1. Browser recording

Microphone Save both browser audio stages and recognition input locally for this comparison

I am fine.
Why are you calling me?
Yes, I make my own financial decisions.
No, please stop calling me.

Press Start, wait one second, read all four phrases with short pauses, wait one second, then Stop. Aim for under 25 seconds. No precise cropping is needed.

0.0 s

Not recording

AudioWorklet input, after browser/OS processing and the existing application low-pass filter. This is not raw hardware audio.

After the same application resampler and PCM16 conversion used by Voice Lab; these samples go to ASR.

2. Make a NEW Windows recording

Do not download or reuse the browser audio above. Record again in the separate Windows Sound Recorder application. The saved browser recording will be reused automatically; do not repeat step 1.

In Windows Sound/Voice Recorder, select the same microphone shown above. Record the same four phrases with surrounding silence, without playing bot audio. Save the file, then select it here. WAV and M4A are supported.

I made a NEW recording in Windows Sound Recorder with the same microphone (not a downloaded browser clip)

These are capture controls, not a listening repair. No model, VAD threshold, echo cancellation, or conversation rules are changed. Recognition runs locally on tiny.en; results remain outside Git.