pietro-guiotto/voice-claude — explained in plain English
Analysis updated 2026-05-18
Give Claude Code instructions by voice instead of typing, using a push-to-talk key.
Navigate Claude Code's menu prompts by speaking numbers aloud.
Use on-device speech transcription to avoid sending audio to external services.
Learn how to wrap a terminal AI tool with a local speech input layer.
| pietro-guiotto/voice-claude | a-bissell/unleash-lite | abhiinnovates/whatsapp-hr-assistant | |
|---|---|---|---|
| Stars | 1 | 1 | 1 |
| Language | Python | Python | Python |
| Setup difficulty | moderate | hard | hard |
| Complexity | 2/5 | 4/5 | 3/5 |
| Audience | developer | researcher | developer |
Figures from each repo's GitHub metadata at analysis time.
macOS only, requires granting microphone, accessibility, and speech recognition permissions.
Voice-claude is a Python wrapper for Claude Code, a terminal-based AI coding assistant, that lets you give instructions by speaking instead of typing. Hold a key, say what you want, release the key, and your words are automatically transcribed and submitted to Claude as if you had typed them yourself. The tool listens for the Right Option key on your keyboard. While held, it records your voice. On release, it uses Apple's Speech Recognition framework to convert the audio to text and types it into Claude. Transcription happens on your device, and no audio is sent to any external service. The tool also tells Claude that input came from speech recognition, so Claude knows to ask for clarification when a transcription sounds unclear rather than acting on a garbled command. Spoken numbers like one, two, three are converted into digits, useful for selecting options in Claude's menu prompts. The README describes this as a prototype and strongly warns against running Claude with full permissions. Transcription errors can cause Claude to receive incorrect instructions, potentially leading to deleted files, force-pushed branches, or unintended commits. Using the most restrictive permission mode available is recommended so destructive actions still require manual approval. The tool runs on macOS only and relies on Apple's Speech Recognition framework and system sounds. It requires Homebrew, Python 3.10 or newer, and Claude Code already installed. Your terminal application needs microphone, accessibility, and speech recognition permissions granted in macOS system settings before it will work. Because it types directly into whatever application has focus, voice-claude is not limited to Claude Code in principle, but the README frames it specifically as a companion for that tool and tests it in that context.
Voice-claude lets you talk to Claude Code by holding a key, speaking, and having your words transcribed on-device and submitted as text.
Mainly Python. The stack also includes Python, macOS Speech Recognition.
License is not stated in the available README content.
Setup difficulty is rated moderate, with roughly 30min to a first successful run.
Mainly developer.
This repo across BitVibe Labs
Verify against the repo before relying on details.