tiny-voca: an 18MB, 4.5M-param TTS model trained without any real human voice, runs on CPU

Dan_Jeffries1 · x · 2026-11-15

Developer @n4ze3m released tiny-voca, a text-to-speech model that has never heard a real human voice during training.

The ultra-lightweight experiment points to a minimal, fully synthetic training approach for on-device speech synthesis.

Related event: tiny-voca: 4.5M-Parameter TTS Trained Fully on Synthetic Speech(4 posts)→

Original post →

More from Multimodal

Multimodal channel →