Audio
Chatterbox
Chatterbox — Resemble AI's open-source TTS family: expressive text-to-speech and voice cloning, multilingual V3, low-latency Turbo and tiny Nano models, MIT
★ 27.1k starsMIT
View guide →MOSS-TTS family: open text-to-speech models for voice cloning, dialogue, voice design, and real-time agents
MOSS-TTS is an open-source audio project by OpenMOSS (OpenMOSS/MOSS-TTS on GitHub). It has 4,184 GitHub stars and 378 forks, is released under the Apache-2.0 license, and was last updated on GitHub on 6 September 2026.
Choose by job: flagship or v1.5 for general TTS/cloning, TTSD for dialogue, VoiceGenerator for designed voices, Realtime for agents, SoundEffect for SFX.
Clone the repo and install with the torch-runtime extra (or the torch-free runtime path).
Fetch the chosen model's weights from Hugging Face or ModelScope as linked in the README.
Run the model's inference example, or serve it behind the OpenAI-compatible speech endpoint.
Flagship, v1.5, local-transformer, dialogue (TTSD), voice-design, realtime, and sound-effect models in one project.
Multilingual synthesis with language tags, explicit [pause X.Ys] control, and Pinyin/phoneme/duration control.
Clone voices from reference audio, with long-form generation support.
MOSS-TTSD targets multi-speaker podcasts, dialogue, and dubbing.
Serve models via SGLang-Omni or vLLM-Omni at /v1/audio/speech for drop-in integration.
PyTorch runtime, optional flash-attention, or a torch-free llama.cpp/ONNX path; fine-tuning tutorials included.
Clone the project and enter its directory.
git clone https://github.com/OpenMOSS/MOSS-TTS.git
cd MOSS-TTSInstall the package with its PyTorch runtime extra (CUDA 12.8 wheels per the README).
pip install --extra-index-url https://download.pytorch.org/whl/cu128 -e ".[torch-runtime]"Download a family member's weights from Hugging Face and follow its README example to synthesize from text (and reference audio for cloning).
# see the chosen model's section in the README for the exact inference commandOpenMOSS, the MOSI.AI lab, maintains the MOSS model family.
Yes — the project is Apache-2.0 licensed and weights are published on Hugging Face and ModelScope.
No. Models run locally; the larger ones expect a capable GPU.
Nano is the tiny 0.1B CPU-friendly sibling. This repo is the full family — up to the 8B flagship — with dialogue, voice design, realtime, and sound-effect models for higher fidelity.
Clone the repo, install with the `.[torch-runtime]` extra, pick a family member, download its weights, and run its README example.
Chatterbox — Resemble AI's open-source TTS family: expressive text-to-speech and voice cloning, multilingual V3, low-latency Turbo and tiny Nano models, MIT
Open-source real-time spoken dialogue with personalized voice cloning.
MOSS-TTS-Nano: a 0.1B open multilingual speech model that clones voices and streams in real time on CPU
Clone a voice from 5 seconds of audio and generate arbitrary speech in real time
VoxCPM2: tokenizer-free TTS for multilingual speech, creative voice design and true-to-life cloning
Multilingual long-form dialogue speech with zero-shot voice cloning.
The most-starred other Audio projects in the LibreAI catalogue — the realistic shortlist if MOSS-TTS is not the right fit.
利用 AI 大模型和自动化工作流，根据主题或关键词一键生成高清短视频。Generate HD short videos from a topic or keyword with an automated AI workflow.
Robust Speech Recognition via Large-Scale Weak Supervision
1 min voice data can also be used to train a good TTS model! (few shot voice cloning)
Clone a voice from 5 seconds of audio and generate arbitrary speech in real time
Port of OpenAI's Whisper model in C/C++
Free and open — grab the source, star the repo, and start creating.