Xiaohongshu open-sources dots.tts.edit V1.1: precise speech editing with 42.92% sequential edit success
小红书技术REDtech · wechat · 2026-11-11
Xiaohongshu's tech team released dots.tts.edit V1.1, an end-to-end precise speech editing model: edit specific words, emotion, pacing or pauses in existing audio while keeping everything else stable, with multiple edits completed in one pass. Supports Chinese/English and multi-speaker audio.
Technical highlights: built on dots.tts's continuous autoregressive architecture, generating in AudioVAE's continuous latent space — no time alignment or duration prediction needed; semi-formal edit instructions bind operations to specific words, with parameters as rate ratios, pitch changes or free-form descriptions.
Results: leads open-source rivals (AuK, Ming-UniAudio, Step-Audio-EditX) on its doteBench; 42.92% full-instruction success on sequential editing, 2x+ the previous best open model; best open-source joint success rates on SpeechEditBench for text, emotion and prosody editing, with prosody beating compared closed-source systems.
The open suite includes weights, inference code, a training data pipeline, the doteBench benchmark, an online editing app and a paper (arXiv:2608.02673).
More from Multimodal
- Local pipeline showcase: MiniMax H3 refMod generations from reference image to 4K upscale — edisson75 · 2026-11-13
- Gaussian splats may be the best way to capture how 3D displays like LKG Portrait look IRL — nptacek · 2026-11-13
- AI-generated 'not canon' clip of Schwarzenegger driving with Michael Jordan — alanskimp · 2026-11-13
- H3-Image API gotcha: server keeps billing on client timeout — set max_retries=0 — xiaohu · 2026-11-13
- MiniMax open-sources H3-Image: native 16MP output, up to 16 reference images — xiaohu · 2026-11-13
- Daily AI painting made with GPT-6 Images 2.6 shows painterly winter village — DeryaTR_ · 2026-11-13