MOSS-TTS-Nano-100M-ONNX

This repository provides the ONNX exports of MOSS-TTS-Nano, a 0.1B multilingual tiny speech generation model from MOSI.AI and the OpenMOSS team. It is designed for torch-free, lightweight deployment on CPU and in the browser, and is intended to be used together with MOSS-Audio-Tokenizer-Nano-ONNX.

Overview

MOSS-TTS-Nano focuses on the part of TTS deployment that matters most in practice: small footprint, low latency, good enough quality for realtime products, and simple local setup. It uses a pure autoregressive Audio Tokenizer + LLM pipeline and keeps the inference workflow friendly for browser demos, local CPU runtimes, and other lightweight integrations.

Main characteristics:

• Tiny model size: about 0.1B parameters
• Native audio format: 48 kHz, 2-channel output
• Multilingual: same language coverage as the PyTorch MOSS-TTS-Nano release
• Pure autoregressive architecture: built on Audio Tokenizer + LLM
• Streaming-friendly export: split into prefill / decode-step / local decoder ONNX graphs
• CPU and browser deployment: designed for onnxruntime and onnxruntime-web

This repository contains the exported ONNX graphs only. If you want the original PyTorch model card and plug-and-play local inference scripts, please use OpenMOSS-Team/MOSS-TTS-Nano or the OpenMOSS/MOSS-TTS-Nano source repository.

Supported Backends

Backend	Runtime	Use Case
ONNX Runtime (CPU)	onnxruntime	Local CPU inference
ONNX Runtime Web	onnxruntime-web	Browser demos / extensions

Repository Contents

File	Description
moss_tts_prefill.onnx	Global transformer prefill graph
moss_tts_decode_step.onnx	Global transformer decode-step graph with KV cache
moss_tts_local_decoder.onnx	Local decoder graph
moss_tts_local_cached_step.onnx	Local cached-step graph
moss_tts_local_fixed_sampled_frame.onnx	Local frame sampling graph
moss_tts_global_shared.data	External weights shared by the global graphs
moss_tts_local_shared.data	External weights shared by the local graphs
tokenizer.model	SentencePiece tokenizer used by the text frontend
tts_browser_onnx_meta.json	Metadata for ONNX runtime integration
browser_poc_manifest.json	Example manifest for browser-based integration

Quick Start

huggingface-cli download OpenMOSS-Team/MOSS-TTS-Nano-100M-ONNX \
    --local-dir weights/MOSS-TTS-Nano-100M-ONNX

huggingface-cli download OpenMOSS-Team/MOSS-Audio-Tokenizer-Nano-ONNX \
    --local-dir weights/MOSS-Audio-Tokenizer-Nano-ONNX

The TTS repo provides the language model and text tokenizer exports, while the companion codec repo provides waveform encode/decode ONNX models.

Main Repositories

Repository	Description
OpenMOSS/MOSS-TTS-Nano	MOSS-TTS-Nano source code, demos, and PyTorch inference
OpenMOSS-Team/MOSS-TTS-Nano	PyTorch MOSS-TTS-Nano weights
OpenMOSS-Team/MOSS-Audio-Tokenizer-Nano-ONNX	Companion ONNX audio tokenizer
OpenMOSS-Team/MOSS-Audio-Tokenizer-Nano	PyTorch audio tokenizer weights
OpenMOSS/MOSS-TTS-Nano-Reader	Browser reading application built on top of the ONNX stack

About MOSS-TTS-Nano

MOSS-TTS-Nano is an open-source multilingual tiny speech generation model built for realtime speech generation and lightweight deployment. The ONNX export keeps the same core architecture as the PyTorch release while making it easier to integrate into browser and CPU-only runtimes without a PyTorch dependency.

For the full project introduction, demos, and PyTorch usage, see:

• MOSS-TTS-Nano Repository
• MOSS-TTS-Nano on Hugging Face

Citation

If you use the MOSS-TTS work in your research or product, please cite:

@misc{openmoss2026mossttsnano,
  title={MOSS-TTS-Nano},
  author={OpenMOSS Team},
  year={2026},
  howpublished={GitHub repository},
  url={https://github.com/OpenMOSS/MOSS-TTS-Nano}
}

@misc{gong2026mossttstechnicalreport,
  title={MOSS-TTS Technical Report},
  author={Yitian Gong and Botian Jiang and Yiwei Zhao and Yucheng Yuan and Kuangwei Chen and Yaozhou Jiang and Cheng Chang and Dong Hong and Mingshu Chen and Ruixiao Li and Yiyang Zhang and Yang Gao and Hanfu Chen and Ke Chen and Songlin Wang and Xiaogui Yang and Yuqian Zhang and Kexin Huang and ZhengYuan Lin and Kang Yu and Ziqi Chen and Jin Wang and Zhaoye Fei and Qinyuan Cheng and Shimin Li and Xipeng Qiu},
  year={2026},
  eprint={2603.18090},
  archivePrefix={arXiv},
  primaryClass={cs.SD},
  url={https://arxiv.org/abs/2603.18090}
}

@misc{gong2026mossaudiotokenizerscalingaudiotokenizers,
  title={MOSS-Audio-Tokenizer: Scaling Audio Tokenizers for Future Audio Foundation Models},
  author={Yitian Gong and Kuangwei Chen and Zhaoye Fei and Xiaogui Yang and Ke Chen and Yang Wang and Kexin Huang and Mingshu Chen and Ruixiao Li and Qingyuan Cheng and Shimin Li and Xipeng Qiu},
  year={2026},
  eprint={2602.10934},
  archivePrefix={arXiv},
  primaryClass={cs.SD},
  url={https://arxiv.org/abs/2602.10934}
}