tiny-voca: 4.5M-param TTS trained on $9.56 of synthetic speech, runs 70x realtime on CPU
aigclink · x · 2026-11-16
Developer n4ze3m released tiny-voca, an English text-to-speech model trained from scratch with only 4.5M parameters, open-sourced under Apache-2.0.
- Trained entirely on synthetic speech from Qwen3-TTS: 17.6 hours / 16,018 clips, total GPU cost just $9.56
- Runs 70x faster than realtime on a 4-thread CPU, outputs 24kHz mono with two US English voices (F2/M2)
- Skips neural vocoders for a Source-Filter architecture: predicts pitch, energy, and voicing features, then synthesizes waveforms via a source signal and dynamic filter
- Ships a custom phonemizer (dizione) that handles numbers, dates, and amounts contextually
English-only for now, no voice cloning; PyTorch and ONNX weights are available for browser/on-device use.
Related event: tiny-voca: 4.5M-Parameter TTS Trained Fully on Synthetic Speech(4 posts)→
More from Multimodal
- LTX 2.5 keeps crashing on AMD 9070 XT with driver timeouts while MiniMax H3 runs fine — ZestycloseReading504 · 2026-11-22
- Stripe residency dev draws all friends in JavaScript with GPT 6.2 Astra, results stun — gabrielchua · 2026-11-22
- Fable 5.2 plus WebGL demoed as a tool to "create anything" — hewarsaber · 2026-11-22
- Cinematic UFO folklore music video built with Suno and SLOPCANNON — DarkroomSignal · 2026-11-22
- Photographer crafts AI-generated digital muse Nous Girl on faux magazine cover — alexcovo_eth · 2026-11-22
- Turn any Google Maps location into a cinematic drone shot with AI — aziz4ai · 2026-11-22