news.volyx.in

Show HN: Kitten TTS – 25MB CPU-Only, Open-Source TTS Model (github.com)

1003 points by divamgupta · 356 days ago · 361 comments on HN

Article summary

Kitten TTS is an open-source, lightweight text-to-speech library that can run on CPU without requiring a GPU, with models ranging from 15M to 80M parameters. The library is built on ONNX and delivers high-quality voice synthesis. It has 8 built-in voices and supports adjustable speech speed and text preprocessing. The library is available for integration into products, with commercial support offered for teams.

Main themes

  • Text-to-Speech Technology
  • CPU-Optimized Models
  • Open-Source Software
  • Voice Synthesis
  • Model Size and Efficiency
  • Commercial Applications

What commenters say

  • The development of small, efficient, and open-source TTS models like Kitten TTS is a promising step towards making AI more accessible and widely adopted.
  • Some commenters believe that dedicated single-purpose hardware with models would be even less energy-intensive, but others argue that this approach may not be economically feasible.
  • There is a debate about the potential for multilingual TTS models, with some arguing that one model per language may be necessary due to the complexities of language and TTS.
  • The quality of Kitten TTS is impressive given its small size, but some commenters note that it may not be the absolute best in quality, and that audio samples are necessary to judge its quality.
  • Some argue that a robotic, explicitly non-natural voice would be perfectly acceptable and even desirable in many situations, while others prefer more natural-sounding voices.
  • The concept of a 'genderless' voice is discussed, with some finding it aesthetically pleasing but others arguing that it is not truly genderless.
  • There is a concern that proprietary models and hardware may lock users into specific ecosystems, while open-source models like Kitten TTS offer more flexibility and freedom.
  • The potential for local inference and edge deployment of TTS models is seen as a key advantage, allowing for more efficient and private use of AI technology.