Show HN: Chirp – Local Windows dictation with ParakeetV3 no executable required (github.com)

Chirp is a new open-source Windows dictation app that runs entirely locally using ParakeetV3 speech-to-text — and it deliberately avoids shipping or requiring .exe installers. If your machine can run Python, you can run Chirp: clone the repo, run the one-time setup to download an ONNX Parakeet bundle, then launch the app with python -m chirp.main (examples use uv run ...). The project bundles nemo-parakeet-tdt-0.6b-v3 (optional int8 quantization), uses ONNX Runtime (default provider "cpu", optionally "cuda" or "dml"), and exposes configurable threading, language auto-detection, post-processing style prompts, clipboard injection, global hotkeys, audio cues, and word-overrides via a simple config.toml. This matters because ParakeetV3 claims near-identical accuracy to Whisper-large-v3 (multilingual WER 4.91 vs 5.05) while being about 17× faster and CPU-only — enabling fast, private, local-first dictation on machines that don’t have GPUs or where installing executables is blocked (common in corporate environments). For practitioners and IT teams, Chirp demonstrates a practical, low-friction path to deploy on-device STT with ONNX and quantization trade-offs, easy customization, and a modular codebase (setup, Parakeet manager, config manager) for extending backends or integrating into workflows.