Credits¶
NarrateClaude's own code is the listener, the shell scripts and the filters listed on the home page. Everything below is someone else's work.
Used by the code in this repo¶
| Project | Used for | By |
|---|---|---|
Speech framework (SFSpeechRecognizer) and AVFoundation |
Speech recognition and mic capture in Listen.swift |
Apple |
Terminal.app AppleScript, osascript, afplay, say |
Window binding, typing, audio playback | Apple |
| Claude Code | The assistant the voice loop drives, the CLAUDE.md project instructions, Stop hooks |
Anthropic |
| MLX | Runs the relevance judge model on Apple Silicon | Apple's ml-explore team |
| mlx-lm | mlx_lm server, the local OpenAI-style endpoint the judge calls |
Apple's ml-explore team |
| Llama 3.2 3B Instruct (4-bit MLX conversion) | Default relevance judge model | Meta, converted by mlx-community |
| Resemblyzer and NumPy | Voiceprint enrollment and matching | Resemble AI, NumPy developers |
Used alongside it¶
| Project | Role | By |
|---|---|---|
| Pocket TTS | The author's cloned voice behind ~/.local/bin/speak |
Kyutai |
| claude-code-local | Runs Claude Code against a local model. Sibling project. | nicedreamzapp |
Named in the 2.0 research logs¶
Not used by the published code: WebRTC audio processing (AEC3, via the PulseAudio webrtc-audio-processing fork), SpeexDSP, FluidAudio, Silero VAD, Smart Turn v3 from Pipecat, and WhisperKit.
This site¶
Built with MkDocs and Material for MkDocs.
If your work is listed here and you want the wording changed, or something is missing, open an issue.