F5-TTS (2024年)（エフファイブティーティーエス）

概要

F5-TTS は2024年10月SWivid + Shanghai Jiao Tong発表・Industry-leading flow matching + Diffusion Transformer 2024年SWivid F5-TTS MIT flow-matching TTS position確立。Industry-leading non-autoregressive + zero-shot voice cloning + 100K hours training + multilingual Chinese+English + MIT license + HuggingFace。

主な特徴・仕組み

• Industry-leading SWivid (Shanghai Wisdom-Information) + Shanghai Jiao Tong University 2024年10月 release
• Industry-leading F5-TTS flow-matching TTS signature
• Industry-leading flow matching signature (vs diffusion sampling)
• Industry-leading Diffusion Transformer DiT signature
• Industry-leading non-autoregressive parallel generation signature
• Industry-leading zero-shot voice cloning signature
• Industry-leading 100K hours training data
• Industry-leading multilingual Chinese + English
• Industry-leading Sway Sampling inference signature
• Industry-leading 24kHz output
• Industry-leading 0.15 Real-Time Factor RTF signature
• Industry-leading MIT license signature
• Industry-leading HuggingFace native + GitHub 9000+ stars at release

まとめ

F5-TTS = 2024年10月SWivid + Shanghai Jiao Tong発表のpremium MIT flow-matching TTS。Industry-leading flow matching vs diffusion sampling + Diffusion Transformer DiT + non-autoregressive parallel generation + zero-shot voice cloning + 100K hours training data + multilingual Chinese + English + Sway Sampling inference + 24kHz output + 0.15 Real-Time Factor RTF + MIT license + HuggingFace native + GitHub 9000+ stars at release + SWivid Shanghai Wisdom-Information + Shanghai Jiao Tong University 2024年10月 release position確立。