1 min read

Kitten TTS v0.8: Lightweight 15M-80M Parameter Text-to-Speech on CPU

Canopy AI

KittenML released Kitten TTS v0.8 with three new open-source ONNX-based text-to-speech models (15M, 40M, 80M parameters) requiring no GPU. Models range from 25 MB to 80 MB on disk and deliver high-quality voice synthesis in 8 voices across Raspberry Pi, smartphones, wearables, and browsers. Targets on-device AI applications without cloud dependency.

Key Takeaways

  • Three new models: 15M parameters (<25 MB int8), 40M parameters (41 MB), 80M parameters (80 MB); int8 quantization support
  • CPU-optimized inference via ONNX; 8 built-in voices; 24 kHz output; speech speed adjustment; runs on Raspberry Pi, mobile, wearables, browsers
  • Production-ready developer preview; multi-lingual release planned; commercial support available through Stellon Labs

Original source: Hacker News / KittenML GitHub