StyleTTS 2
Open TTS with style diffusion that aimed at human-level naturalness.
About
StyleTTS 2 is an open text-to-speech model that models speaking style as a latent variable through diffusion, so it can pick a fitting style for the text without needing a reference clip. The authors reported it beating human recordings on the single-speaker LJSpeech set in listener tests and matching them on multi-speaker VCTK. The code is MIT-licensed, though the pre-trained voice models carry their own use terms around voice consent. It is showing its age next to newer zero-shot cloners, and setup is aimed at people comfortable training and running research code.
Specs
Context window
Pricing
Free quota
API access
Weights
Released
Benchmarks
Authors reported surpassing human recordings on LJSpeech and matching them on VCTK in native-listener tests (paper, 2023).
Provider-reported unless a source says otherwise. Benchmarks are marketing-adjacent by nature — treat them as directional.
Use cases
Trust score — 7/10
Scored 7/10 (verified tier, metrics-v1). Evidence: 6.3k GitHub stars on yl4579/StyleTTS2; widely referenced in TTS research; Repo quiet since ~2024; no new releases, community forks carry it forward; Code is MIT; pre-trained voice models carry consent/disclosure use terms; Human-level claims are from the authors' 2023 paper listener tests, not independent replication. License: Code MIT per repo; pre-trained model use requires voice-consent disclosure per the README..
How the trust score is built
Full methodology →- Adoption · stars/downloads, log-scaled vs peers
- 7.0/10
- Maintenance · recency, releases, issue load
- 5.0/10
- Openness · license + docs
- 8.0/10
- Manual review · sanity read + red-flag check
- 7.0/10
- · 6.3k GitHub stars on yl4579/StyleTTS2; widely referenced in TTS research
- · Repo quiet since ~2024; no new releases, community forks carry it forward
- · Code is MIT; pre-trained voice models carry consent/disclosure use terms
- · Human-level claims are from the authors' 2023 paper listener tests, not independent replication