Skip to content

auto-speech v0.1.0

Latest
Latest

Choose a tag to compare

View all tags
@JoshuaRamirez JoshuaRamirez released this 22 Jul 06:17
v0.1.0
05cc00a

First public release of auto-speech — Claude Code, out loud.

Highlights

Playback engine

  • Local Kokoro-82M TTS on Apple Silicon (mlx-audio) — no cloud TTS, no API keys
  • Fibonacci-scaled buffered playback through mpv: sub-second time-to-first-audio on long responses
  • Source-hash replay cache; /auto-speech-replay re-speaks instantly

End-of-turn autoplay

  • A Claude Code Stop hook reads each completed response aloud — on by default, per-session opt-out
  • verbatim / small / medium / large read modes; all/solo session scope
  • Cross-session FIFO queue with coalescing and dedup — an active playback is never cut off

Web app + Chrome extension

  • Localhost web app (127.0.0.1:7860): paste-to-speak with Claude-CLI rewrite, transport controls, replay browser, multi-voice
  • AutoSpeech Chrome extension (v0.4.0, attached below): right-click → speak selection, backed by the local server with browser-voice fallback

Real-time narrator (optional)

  • Hook events → phase classifier → local MLX LLM → one spoken sentence per phase transition, while Claude works

Install

Apple Silicon Mac + Homebrew + uv required.

git clone https://github.com/JoshuaRamirez/auto-speech && cd auto-speech
bash setup/install.sh          # venv + locked deps + mpv + jq
bash setup/verify.sh           # you should hear audio
bash setup/install-plugin.sh   # 8 slash commands
bash setup/install-hook.sh     # the autoplay Stop hook

Full quickstart and options in the README.

Security

Local-first by design: synthesis and narration run entirely on-device; the server binds loopback and grants CORS solely to Chrome-extension origins; hooks are installed transparently and the optional self-update runs uv sync against the committed lockfile only. The rewrite step that shapes text for speech goes through your own claude CLI login — the same trust boundary as typing into Claude Code. Full threat model and reporting: SECURITY.md.

Docs

README · Operations runbook · Contributing · Design history (OOAD spec, 14 ADRs, per-phase designs)

Assets

  • auto-speech-chrome-extension-v0.4.0.zip — the AutoSpeech extension, built from this tag. Unzip, then chrome://extensions → Developer mode → Load unpacked. Requires the local web app running for Kokoro-quality audio (falls back to browser voices otherwise).

Full change list: CHANGELOG.md

Assets 3
Loading