#6262nd of 67 in VoiceFor voice, speech and music
megatts3
Zero-shot Chinese and English voice-cloning TTS with a 0.45B diffusion transformer
- Stars
- 6.1k
- License
- Apache-2.0
- Last commit
- Jun 2026
- Language
- Python
6,002 on 3 Apr 20266,118 on 30 Oct 2026
From GitHub's daily star counts.
Checked on GitHub.
Overview
MegaTTS 3 is ByteDance's PyTorch implementation of a latent diffusion transformer for zero-shot speech synthesis in Chinese, English and code-switched text. It ships a CLI and a Gradio web UI, plus an aligner, a Qwen2.5-0.5B grapheme-to-phoneme model and a WaveVAE that can be used on their own. The WaveVAE encoder is not released, so cloning needs pre-extracted .npy latents for each voice.
Who it is for: Speech researchers testing zero-shot voice cloning in Chinese and English
Strengths
- Diffusion Transformer backbone is only 0.45B parameters
- Accent strength adjustable through p_w and t_w inference weights
- Aligner, G2P and WaveVAE submodules are reusable outside TTS
- CPU inference works, about 30 seconds for 10 steps
Weaknesses
- WaveVAE encoder withheld; custom voices need .npy latents supplied by the authors
- Only Chinese and English supported
- Windows and Docker setups are marked as under testing
- Fine-grained pronunciation and duration control is listed as coming soon
What it needs
- GPU optional
- Docker
- Needs Python 3.10, PyTorch 2.6.0, Gradio, ffmpeg, WeTextProcessing
- Models: MegaTTS3 checkpoints (Hugging Face, Google Drive), Qwen2.5-0.5B G2P, WaveVAE
- port 7860
Quick answers
Is megatts3 open source?
Yes. Its code is on GitHub under the Apache-2.0 license.
Can you run megatts3 with Docker?
Yes. Its repository has a Docker setup: a Dockerfile, a Compose file or a published image.
Does megatts3 need a GPU?
No. It runs on a CPU and can use a GPU when you have one.
Also in Voice
See all 67| Rank | Project | Score |
|---|---|---|
| 1 | voicestudioDesktop app for voice cloning, dubbing, dictation and audiobooks | 97 out of 100 |
| 2 | whisper.cppC/C++ port of OpenAI Whisper for local speech-to-text transcription | 89 out of 100 |
| 3 | handyOffline push-to-talk dictation app that pastes transcripts into any text field | 89 out of 100 |
| #4 | xiaozhi-esp32ESP32 firmware for a voice chatbot with MCP device and cloud tool control | 82 out of 100 |
| #5 | buzzOffline desktop app for transcribing and translating audio with Whisper | 81 out of 100 |