jmn-setup-whisper

OpenAI Whisper — voice input for Claude Code sessions. ## Installation - **Package:** `python-openai-whisper` (Arch `pacman`) - **Version:** 20250625-3 - **Binary:** `/usr/bin/whisper` - **Install date:** 2026-03-23 ## Voice Commands Scripts at `~/.local/bin/v1`-`v5` — record mic via PipeWire, transcribe with Whisper, output text to stdout + clipboard. | Command | Duration | |---------|----------| | `! v1` | 10s | | `! v2` | 20s | | `! v3` | 30s | | `! v4` | 40s | | `! v5` | 50s | ### Continuation keyword If transcription ends with "TBC" (to be continued), the message is incomplete — wait for the next voice input before responding. ### Config (env vars) - `WHISPER_MODEL` — model size (default: `base`) - `WHISPER_LANG` — language (default: `en`) ## Notes - Default model is `base` (139 MB) — good balance of speed and accuracy - Runs on CPU by default; add `--device cuda` for GPU acceleration - Dependencies managed by pacman (pytorch, numba, numpy, etc.)