Show HN: Greek and Cypriot Greek TTS that runs faster than real time on CPU
Category: ai-ml
Tags: text-to-speech, cpu-inference, voice-cloning, streaming, machine-learning
Score: 7.5/10 (Innovation: 7, Technical: 8, Documentation: 8, Utility: 7)
Pocket TTS is a lightweight text-to-speech system by Kyutai that runs efficiently on CPUs, achieving faster-than-real-time synthesis (~6x on a MacBook Air M4) with only 2 cores. It supports streaming, voice cloning, and multiple languages, and can be used via CLI, Python API, or even in the browser. The project is notable for its compact 100M parameter model, low latency (~200ms first chunk), and MIT license.
Target audience: developers, researchers, and hobbyists needing offline TTS on CPU, including backend devs, embedded systems developers, and accessibility tool creators.
Repository: https://huggingface.co/myned-ai/pocket-tts-greek · Python · MIT · 9990 stars
View on Hacker News