LongCat-AudioDiT-1B: A high-fidelity waveform latent diffusion text-to-speech model
Original title:meituan-longcat/LongCat-AudioDiT-1B
AI summary
The Meituan LongCat team's open-source diffusion TTS model abandons the traditional mel-spectrogram intermediate representation and operates directly in waveform latent space, synthesizing speech with only a Wav-VAE and diffusion backbone. The model fixes the training-inference mismatch and replaces classifier-free guidance with adaptive projection guidance. The largest version, 3.5B, achieves SOTA zero-shot voice cloning on the Seed benchmark, with speaker similarity (SIM) reaching 0.818 on Seed-ZH and 0.797 on Seed-Hard, surpassing the previous best Seed-TTS. The study also finds that Wav-VAE reconstruction fidelity is not positively correlated with final synthesis quality.
Key Takeaways
- The Meituan LongCat team's open-source diffusion TTS model abandons the traditional…
- The model fixes the training-inference mismatch and replaces classifier-free guidance with…
- The largest version,3.
- 5B,achieves SOTA zero-shot voice cloning on the Seed benchmark,with speaker similarity (SIM)…
- 818 on Seed-ZH and 0.
Quick Facts
- Published
- Updated
- Source
- 美团 LongCat:HuggingFace 新模型
- Category
- Other
美团开源 1B 语音克隆模型,Seed 基准超 Seed-TTS,零样本推理可用
Source & copyright
This article is based on curated titles and AI summaries from the AIHOT public API, with attribution and links back. Original copyright remains with the original publishers and authors; this site does not mirror full texts.
How to Cite
When citing this article, keep the title, date, and permalink, and credit Orange News as the source.
Orange News: LongCat-AudioDiT-1B: A high-fidelity waveform latent diffusion text-to-speech model. March 30, 2026. https://new.nat6.net/en/news/cmocymrfh0057slqlbzifa44t/