Show HN: Whisper Menu Bar – a push-to-talk transcription script (gist.github.com)

Show HN: Whisper Menu Bar is a minimal macOS menu-bar push-to-talk transcription script that runs OpenAI’s Whisper locally. Built with PyObjC, PyAudio and the openai-whisper package, it installs as an accessory app (no Dock icon), monitors the Option key globally to start/stop recording, saves captured audio to a temporary WAV, and transcribes it in a background thread. The transcript is copied to the macOS clipboard and the menu shows model selection (tiny, base, small, medium), status updates, and a red recording indicator. Audio settings: mono, 16 kHz, 16-bit, 1024-frame chunks; transcription uses fp16=False and loads models on CPU by default (CUDA if available). Significance: this is a compact, hackable example of local speech-to-text integration that prioritizes speed of use and privacy (local inference) over cloud services. It demonstrates real-world patterns—background model loading, thread-safe UI updates via NSOperationQueue, pyaudio stream callbacks, and temp-file based transcription—that other developers can borrow for ephemeral note-taking, accessibility tools, or clipboard workflows. Caveats: macOS-only due to AppKit/PyObjC, CPU inference on larger models can be slow or memory-heavy, and the script assumes microphone permissions and PyAudio drivers are available.