← sources

tiny-voca: a fully open 4.5M-parameter TTS (code, weights, dataset). Runs 70x real time on a CPU

source: reddit (reddit:LocalLLaMA)published 2026-11-14 17:26ingested 2026-11-15 02:36

retained text

I made a small English text-to-speech model called tiny-voca. It has 4.5M parameters, two voices, and it runs in your browser.  


Try it here, nothing is sent to a server: [https://tiny-voca.n4ze3m.com](https://tiny-voca.n4ze3m.com)

Some facts:

\- 4,537,291 parameters, 18 MB. This is the whole model, the waveform part is included.  
\- Two voices, one female and one male. 24 kHz.  
\- About 70x faster than real time on 4 CPU threads.  
\- Runs with PyTorch, ONNX, and in the browser.  
\- Trained from scratch in 7.7 hours on one RTX 4090.

What I think is different:

\- Everything is open. Code, weights, the training recipe and the dataset.  
\- Custom phonemizer.  
\- No neural vocoder. The sound is made by a small source-filter synth, which is why it is fast.  


What is not good yet:

\- English only, and only these two voices. No voice cloning.  
\- Foreign names and odd spellings are often read wrong.  
\- It can sound a bit synthetic.  
\- The training audio is synthetic too. I made it with Qwen3-TTS, so the model can't sound better than that.

Links:

\- Demo: [https://tiny-voca.n4ze3m.com](https://tiny-voca.n4ze3m.com)  
\- Code: [https://github.com/n4ze3m/tiny-voca](https://github.com/n4ze3m/tiny-voca)  
\- Weights: [https://huggingface.co/n4ze3m/tiny-voca](https://huggingface.co/n4ze3m/tiny-voca)  
\- Dataset: [https://huggingface.co/datasets/n4ze3m/tiny-voca](https://huggingface.co/datasets/n4ze3m/tiny-voca)  
\- Phonemizer: [https://github.com/n4ze3m/dizione](https://github.com/n4ze3m/dizione)  


Please try hard text: numbers, dates, names, long sentences. If something sounds wrong, post the text and the voice. That helps me a lot.

document fetched 2026-11-15 00:36 · canonical https://reddit.com/r/LocalLLaMA/comments/1xw7zrg/tinyvoca_a_fully_open_45mparameter_tts_code