Kani TTS 370M Offers Compact Multilingual Speech

A new, efficient text-to-speech model called Kani TTS 370M has been released on the Hugging Face Hub. Developed by the user nineninesix, the model contains 370 million parameters, offering a relatively lightweight option for generating high-quality, multilingual speech.

The model is based on the Language-Free Modeling for Multilingual Text-To-Speech (LFM2) architecture. This approach allows it to handle multiple languages without relying on explicit language identification tags during training or inference. This design choice can make a model more flexible and scalable for diverse linguistic applications, learning to synthesize different languages from a mixed dataset.

Kani TTS 370M's compact size is its most notable feature. In a field often dominated by multi-billion parameter models, a smaller footprint makes it more accessible for researchers and developers with limited computational resources. This could enable its use in on-device applications or lower-cost cloud deployments where efficiency is a primary concern.

The model weights and usage instructions are publicly available on its Hugging Face repository. While the weights are accessible, the license is listed as "All rights reserved," indicating that it is not intended for commercial use without permission from the creator.