MOSS-TTS-Nano Delivers Multilingual Speech at 100M Params

The field of open-source text-to-speech has a new, compact contender. A group known as OpenMOSS-Team has released MOSS-TTS-Nano, a generative audio model with just 100 million parameters designed for high-quality, multilingual speech synthesis.

The model's key feature is its linguistic flexibility. It officially supports English, Mandarin Chinese, and Cantonese, but its most notable capability is handling mixed-language sentences—a common challenge for speech models. This allows it to generate natural-sounding audio from text that switches between languages, such as "give me a cup of 拿铁."

At just 100M parameters, the 'Nano' in its name is well-earned. This small size makes MOSS-TTS-Nano a compelling option for applications where computational resources are limited, such as on-device assistants, embedded systems, or other edge computing scenarios. It presents an efficient alternative to larger, cloud-dependent text-to-speech APIs.

The model is available for download from the team's Hugging Face repository. It's released under a Creative Commons CC BY-NC-SA 4.0 license, which permits academic and personal use but restricts commercial applications.