MeloTTS
MyShellA multilingual text-to-speech library from MyShell and MIT, with voices for several English accents, Spanish, French, Chinese mixed with English, Japanese and Korean, fast enough for real-time synthesis on a CPU.
The repository has not been pushed to since December 2024 and the newest checkpoint dates from April 2024. A PyPI package named melotts exists but names no repository, so it is not declared.
Openness
3 high confidence- weights
- open(eight per-language .pth checkpoints on the Hub, ungated)
- data
- closed(no statement of what the released checkpoints were trained on)
- code
- open(training scripts in the repository (train.sh, preprocess_text.py))
- license
- MIT(code and every checkpoint)
MeloTTS is MIT-licensed, code and checkpoints alike, and the repository includes the scripts to train a voice on your own data. What the released voices were trained on is not stated, so they cannot be rebuilt.
- https://huggingface.co/api/models?author=myshell-ai&search=MeloTTS&limit=100 recorded 2026-09-27
Family listing: eight myshell-ai/MeloTTS-* checkpoints tagged "license:mit", the newest MeloTTS-English-v3 created 2024-04-17.
- https://huggingface.co/api/models/myshell-ai/MeloTTS-English recorded 2026-09-27
Hub record: "gated":false and checkpoint.pth in the file list.
- https://raw.githubusercontent.com/myshell-ai/MeloTTS/HEAD/docs/training.md recorded 2026-09-27
Training doc: "To train a TTS model, we need to prepare the audio files and a metadata file."; "bash train.sh <path/to/config.json> <num_of_gpus>".
- https://raw.githubusercontent.com/myshell-ai/MeloTTS/HEAD/LICENSE recorded 2026-09-27
LICENSE body: "Copyright (c) 2024 MyShell.ai" followed by the MIT permission text.
- https://raw.githubusercontent.com/myshell-ai/MeloTTS/HEAD/README.md recorded 2026-09-27
README: "This library is under MIT License, which means it is free for both commercial and non-commercial use."
Adoption
3 high confidenceHugging Face downloads summed over the per-language checkpoints, about half of them the English voice.
- https://huggingface.co/api/models?author=myshell-ai&search=MeloTTS&limit=100 recorded 2026-09-27
Trailing-30-day downloads: MeloTTS-English 263,685, English-v3 62,843, Korean 62,528, Japanese 54,079, Chinese 37,345, Spanish 14,047, French 8,055.
Capability
1 low confidenceA light, practical synthesizer for fixed voices. It clones no voices and publishes no quality results, and listeners have not rated it on the public arena, so it sits well below Kokoro.
- https://artificialanalysis.ai/text-to-speech/leaderboard recorded 2026-09-27
The 91-row leaderboard has no MeloTTS entry.
- https://raw.githubusercontent.com/myshell-ai/MeloTTS/HEAD/README.md recorded 2026-09-27
README: "MeloTTS is a **high-quality multi-lingual** text-to-speech library"; "Fast enough for `CPU real-time inference`."
Verified 2026-09-27