MOSS-TTS
LibraryOpenMOSS/MOSS-TTS
Open-source speech and sound generation model family for high-fidelity TTS.
Overview
MOSS-TTS Family provides a suite of models for text-to-speech, voice cloning, sound effects, and real-time streaming. It supports multilingual synthesis and fine-tuning, with backends for efficient inference. Designed for high-expressiveness and complex real-world scenarios.
README Preview
# MOSS-TTS Family\n\n\n\n\n \n \n \n\n\n\n\n\n \n \n \n \n \n\n \n \n \n \n \n\n\n\n[English](README.md) | [简体中文](README_zh.md)\n\n\nMOSS‑TTS Family is an open‑source **speech and sound generation model family** from [MOSI.AI](https://mosi.cn/#hero) and the [OpenMOSS team](https://www.open-moss.com/). It is designed for **high‑fidelity**, **high‑expressiveness**, and **complex real‑world scenarios**, covering stable long‑form speech, multi‑speaker dialogue, voice/character design, environmental sound effects, and real‑time streaming TTS.\n\n\n## News\n* 2026.5.26: 🚀 Released [MOSS-SoundEffect-v2.0](https://huggingface.co/OpenMOSS-Team/MOSS-SoundEffect-v2.0), a new text-to-audio model using a **DiT backbone with the Flow Matching objective**, generating **48 kHz** bilingual sound effects up to **30 seconds** — see [`moss_soundeffect_v2/`](https://github.com/OpenMOSS/MOSS-TTS/tree/main/moss_soundeffect_v2).\n* 2026.5.26: 🚀 Released [MOSS-TTS-v1.5](https://huggingface.co/OpenMOSS-Team/MOSS-TTS-v1.5), with stronger multilingual synthesis when language tags are provided, more stable voice cloning, better long-reference short-text cloning, punctuation-following prosody, and explicit pause control via `[pause X.Ys]`.\n* 2026.5.6: 🚀 MOSS-TTS and MOSS-Audio-Tokenizer now support `mlx-audio`. Visit the [mlx-audio GitHub repository](https://github.com/Blaizzy/mlx-audio) for details.\n* 2026.4.29: 📝 MOSS-TTS 2.0 is coming soon! We are collecting TTS feedback, suggestions, and feature requests via the [requirements collection form](https://acnc6zeentra.feishu.cn/share/base/form/shrcnyAe1LwqKWjCSuW4wiZ2Hef).\n* 2026.4.13: 🚀 MOSS-TTS-Nano, our ~100M-parameter model, is now available! It supports multilingual voice cloning, 48 kHz stereo input/output, and streaming output on just 4 CPU cores. Check the [GitHub repository](https://github.com/OpenMOSS/MOSS-TTS-Nano) and our [blog](https://openmoss.github.io/MOSS-TTS-Nano-Demo/) for more detai
FAQ (3)
TroubleshootingHow to fix the /api/v1/voices public voice API endpoint returning no voice data?
This issue was caused by an abnormal backend service configuration. The official team has completed the production deployment of the fix, and the /api/v1/voices endpoint now returns voice data normally. Please retry your API request. If the problem persists after retry, contact official support for further assistance.
TroubleshootingHow to prevent MOSS-TTS v1.5 Continuation + Clone from repeating the reference transcript instead of the Japanese target text?
This occurs because mode="continuation" treats the reference audio and ref_text as an actual prefix and generates speech that continues from them. When the reference and target texts are unrelated, the model can repeat the prefix, substitute words from it, or insert it before the target text.
For short utterance voice cloning with unrelated text, use Clone mode instead of Continuation + Clone. Pass only the target text and reference audio, and omit the assistant prefix audio:
conversations = [[
processor.build_user_message(
text=text_2,
reference=[ref_audio_2],
language="ja"
)
]]
batch = processor(conversations, mode="clone")Then run the same model.generate(...) and decoding flow.
Use mode="continuation" only when the reference audio is genuinely the preceding speech and the target text is its natural continuation. Do not concatenate an unrelated reference transcript with the target text, as in text=ref_text_2 + text_2, when you only want timbre cloning.
TroubleshootingWhy is the pause tag not working in Mosi/MOSS TTS Studio?
The model deployed in the Studio experience may differ from the documentation. Try the latest model at https://platform.mosi.cn/, where MOSS TTS 1.5 is deployed; users reported improved pause-tag and instruction-following behavior. Pinyin tone control such as zuo4 cheng2 shi4 is also more reliable there, but generation can still be inconsistent, so regenerating the audio may be necessary. There is no fully deterministic fix in the reported version.