Perch Audio

Microphone capture for Perch dictation.

Perch runs where your workspace is. Over Remote-SSH that is another machine, which cannot reach your microphone. Perch Audio runs on the computer you sit at, records when Perch asks it to, and either hands the recording over for Perch to turn into text, or turns it into text itself and hands over the words. It does the latter when this computer has an NVIDIA GPU (or perch.voice.runOn is local), with the same Whisper engine Perch has, set up once into ~/.local/share/perch/voice here.

It is installed with Perch. On its own it does nothing.

• Perch Audio: Choose Microphone… picks the microphone to record from. Bluetooth headsets are listed even while in their music profile; one is switched to headset mode for a recording and back after.
• perchAudio.device names it; empty uses the system default.
• perchAudio.maxSeconds is how long a recording may run, 180 by default.
• perchAudio.saveRecordings, a diagnostic: a directory on this computer where each recording is saved as a WAV, with a transcripts.log of what was heard when the words are worked out here. Empty, the default, saves nothing.

Audio is held in memory only while a recording is in progress. Nothing is written to disk, and nothing here reads workspace files.