[connorjfarrell]

Adds Whisper 0.1.0: it transcribes recorded calls with a Whisper server (whisper.cpp's whisper-server) and writes each transcript beside its call as <call>.transcript.json.

• Choice per system and talkgroup: each system says whether to transcribe, which talkgroups, and which server (so which model) each kind of talkgroup uses, by number or by words in its name or category. A fast model for most, a better one for the talkgroups that matter.
• Privacy: audio goes only to the servers you name in its settings. Calls the recorder decrypted are given to it only if you turn on Share decrypted calls; calls that could not be decrypted are never transcribed.
• Catching up: new calls are transcribed newest first, and earlier calls can be caught up on in the background. Helper servers on other machines (a Mac mini, say) take the newest waiting call when free. server/ has the Docker (Vulkan) and Apple Silicon (Metal) setups, tuned on real radio audio (a short audio window and --no-context matter a lot).
• Checked: ./check plugins/whisper.json passes; the release workflow built all four platforms with attestations. 47 tests, clippy -D warnings and fmt are clean. I run it on a live recorder, with thousands of real calls transcribed.

Playback and Retention know about the transcript files (Retention deletes them with their call).

🤖 Generated with Claude Code