MOSS-TTS-Nano

OpenMOSS's 100M text-to-speech model for CPUs, with voice cloning, around 20 languages and 48 kHz stereo output. Apache 2.0.

In short. MOSS-TTS-Nano is a very small voice model that runs on an ordinary computer's processor, with no graphics card. It reads text in around 20 languages, copies a voice from a short recording, and starts speaking while it is still generating. It is free to use in a business.

Summary

MOSS-TTS-Nano is a tiny text-to-speech model of about 100M parameters from MOSI.AI and the OpenMOSS team, released on 10 April 2026. It follows the audio-tokenizer-plus-language-model recipe of the larger MOSS-TTS in a package built for CPUs. The README says it writes 48 kHz stereo audio, supports 20 languages (its table lists 19, and Hebrew from the MOSS-TTS list is missing), streams its output with a short time to first audio, splits long text into chunks automatically while it clones a voice from a reference clip, and that streaming generation can run on a 4-core CPU. The table lists Chinese, English, German, Spanish, French, Japanese, Italian, Hungarian, Korean, Russian, Persian, Arabic, Polish, Portuguese, Czech, Danish, Swedish, Greek and Turkish.

An ONNX version, added on 17 April 2026, drops the PyTorch dependency. OpenMOSS says it nearly doubles processing efficiency in its tests, and that it ran smoothly on a single CPU core of a MacBook Air M4. The repository also has a local web demo, a command line tool with a server mode, fine-tuning code and an Android ONNX Runtime example. A companion project, MOSS-TTS-Nano-Reader, runs the model inside the browser as an extension. The model uses the small MOSS-Audio-Tokenizer-Nano codec, and the project publishes reconstruction scores for that codec but no word-error-rate or speaker-similarity benchmark for the speech model itself, so judge the voices by listening to the demo.

The weights are Apache 2.0 and the PyTorch checkpoint is 0.24 GB. The README's own license section is a placeholder, but the repository has an Apache 2.0 LICENSE file and the Hugging Face page lists the same license. In the 30 days to 11 November 2026, Hugging Face counted about 119,000 downloads. For stronger voices in more languages, the same team's MOSS-TTS is listed as a separate model here.

Good for

  • Text to speech on a laptop, a mini PC or a small server with no GPU.
  • Offline and in-browser readers, since the project ships a browser extension and an ONNX route that avoids PyTorch.
  • Prototyping voice cloning in many languages with a 0.24 GB model.
  • Android experiments, using the ONNX Runtime example in the repository.

Watch out for

  • The project publishes no listening or error-rate benchmark for the speech model, only codec reconstruction scores.
  • The README says voice cloning is the main recommended workflow, so check the built-in voices yourself before you rely on them.
  • Quality at this size will not match the 8B MOSS-TTS or other much larger models, and Cantonese, Dutch, Thai and Vietnamese are not in its language table.
  • The README's license section is a placeholder, so rely on the LICENSE file, and clone a real person's voice only with consent.

Performance claims come from the developer and have not been independently verified.

How much memory does it need?

Memory figures for each variant of MOSS-TTS-Nano. Image, video, speech and music pipelines also load a text encoder and a decoder, so the memory needed can be larger than the main checkpoint.

Variants Parameters Download size Memory needed Fits on
Nano (100M) 100M 0.24 GB 0.276 GB Laptop or phone

Download size covers every component needed to run the model, such as the generator, text encoder and decoder. Memory needed is a stated or published measured minimum where one exists; otherwise it is the download size plus about 15%. Larger images, longer clips and higher resolutions need more.

The license in plain words

One of the most widely used permissive licenses. You may use, change and sell products built on the model with only light conditions.

More about Apache 2.0

See also

Frequently asked questions

What is MOSS-TTS-Nano?
MOSS-TTS-Nano is a very small voice model that runs on an ordinary computer's processor, with no graphics card. It reads text in around 20 languages, copies a voice from a short recording, and starts speaking while it is still generating. It is free to use in a business.
Can I use MOSS-TTS-Nano commercially?
Yes. MOSS-TTS-Nano is released under Apache 2.0, a permissive license. You can use it commercially if you follow the standard conditions, such as keeping the license notice.
Is MOSS-TTS-Nano open source?
MOSS-TTS-Nano is released under Apache 2.0, a license approved by the Open Source Initiative. Training data and training code may not be public, so it is best described as open weights under an open-source license.
Can I run MOSS-TTS-Nano on my own computer?
Yes. The smallest variant, Nano (100M), needs about 0.276 GB of memory at 4-bit precision, which fits a modern laptop or a mid-range graphics card.
When was MOSS-TTS-Nano released?
MOSS-TTS-Nano was released by OpenMOSS on April 10, 2026.

Official links

Sources

Specifications come from these primary sources. Benchmark claims, where mentioned, are reported by the developer.

Release date and facts last verified November 11, 2026.

Related models

Model Released Parameters Context License Smallest variant needs
MOSS-TTS OpenMOSS 4.55B – 8.49B – Apache 2.0 (Permissive) 10 GB Laptop or phone
Kokoro hexgrad 82M – Apache 2.0 (Permissive) 0.38 GB Laptop or phone
Pocket TTS Kyutai 18M – 100M – CC BY 4.0 (Permissive) 0.127 GB Laptop or phone
Phonon 31M Gradium 31M – Apache 2.0 (Permissive) 0.0747 GB Laptop or phone