# StyleTTS 2

> Research code for the StyleTTS 2 paper (arXiv 2306.07691). Code is MIT; the pretrained checkpoints on Hugging Face carry no licence file, only usage terms in the README requiring you to disclose synthetic speech or have the speaker's permission. Inference runs from Jupyter notebooks and depends on a GPL-licensed phonemizer, which is why it is not bundled. English-focused (the PL-BERT module is English-only). Inference hardware is not documented. Last push to the repo was August 2024.

- Repository: https://github.com/yl4579/StyleTTS2
- Official docs: https://github.com/yl4579/StyleTTS2#readme
- Checked: 2026-10-15

## What it replaces
- ElevenLabs: Speech generation, Voice cloning (partial). Research-grade English TTS with zero-shot speaker adaptation from the LibriTTS model; inference is via notebooks, not a ready app, and other languages need a different PL-BERT.
- Descript: Voice cloning, Speech generation (partial). Voice cloning and text-to-speech only; no transcript-based audio or video editing.

## Facts, with sources

| Question | Status | Answer | Source | Checked |
|---|---|---|---|---|
| Runs on my machine? | Confirmed | Yes, Python/PyTorch inference via the provided notebooks | https://github.com/yl4579/StyleTTS2#inference | 2026-10-15 |
| Code licence | Confirmed | MIT | https://github.com/yl4579/StyleTTS2/blob/main/LICENSE | 2026-10-15 |
| Model weights licence | Partial | No licence on the Hugging Face model repos; README usage terms require disclosing that speech is synthetic unless you have the speaker's permission | https://github.com/yl4579/StyleTTS2#inference | 2026-10-15 |
| Commercial use | Partial | Code is MIT, but the pretrained weights have no formal licence (only disclosure/permission terms) and inference relies on a GPL-licensed phonemizer; an MIT-only package using gruut exists at lower quality | https://github.com/yl4579/StyleTTS2#inference | 2026-10-15 |
| Paid services needed? | Confirmed | None required | https://github.com/yl4579/StyleTTS2#inference | 2026-10-15 |
| Hardware | Partial | Inference requirements not documented. CPU inference is suggested as a fix for noise on older GPUs. Fine-tuning example took about 4 hours on four NVIDIA A100s. | https://github.com/yl4579/StyleTTS2#common-issues-2 | 2026-10-15 |
| Operating systems | Partial | Linux install steps (apt espeak-ng) and an extra Windows PyTorch step are given; macOS not mentioned | https://github.com/yl4579/StyleTTS2#pre-requisites | 2026-10-15 |
| Setup effort | Confirmed | Medium. Clone, pip install requirements, install phonemizer and espeak-ng, download checkpoints, run notebooks | https://github.com/yl4579/StyleTTS2#pre-requisites | 2026-10-15 |

GitHub: 6363 stars, 709 forks, last push 2024-08-10, archived: no.
---
Source: RepoQuarry (https://repoquarry.dev/projects/styletts2). Data licence: CC BY 4.0. Quotes are from project documentation and belong to their authors.