apple-llm adds Mac-native TTS/STT: 30 min narration renders in under 20s on M4
light_2earth · reddit · 2026-11-06
apple-llm, a Node/Python package that calls macOS's built-in LLMs without an API key, now adds system voices and Apple speech recognition in v0.3 — fully local.
Measured on an M4 MacBook Air:
- 30 minutes of narration renders in 18–20 s using 50 MB RAM; 60 minutes in 40 s
- 30 minutes transcribed in 28 s with optional per-word timestamps; 3–4% WER on clean speech
- 184 voices and 68 recognition languages available
Usable via CLI (speak/transcribe/listen) or Python; the local server exposes OpenAI-compatible /v1/audio/speech and /v1/audio/transcriptions, so swapping the base URL makes any OpenAI client work. Includes a 20-line voice assistant example. Limits: no Siri voices, no MP3/Opus output. Requires macOS 26+ on Apple Silicon, MIT licensed.
More from coding & agent
- SpaceX AI engineer runs a team of 15 AI assistants: 1 chief of staff, 3 managers, 11 workers — JarnoDuursma · 2026-11-06
- Engineers' focus time is shrinking — here's how to reestablish flow at work — rseroter · 2026-11-06
- SWE-bench author responds to benchmaxx concerns: SWE needs human-in-the-loop benchmarks — OfirPress · 2026-11-06
- Qwen 3.8 Flash tested on an RTX 3080 Ti: builds an Uber demand simulation end to end — dreamai87 · 2026-11-06
- Clawdock is an open-source Windows tool that turns your Copilot key into a Claude voice assistant — Necessary-Aioli-8402 · 2026-11-06
- How SpaceXSI uses Grok Bot as a proactive GTM teammate, part 2 — dean_rie · 2026-11-06