Tencent open-sources AuK: a 1.5B speech model that generates and edits audio via text prompts
realmrfakename · x · 2026-09-10
Tencent released AuK, an open-source 1.5B-parameter foundational model that unifies speech generation and editing through natural-language instructions, with weights fully available.
- Five task families: speech generation, content editing, enhancement/separation, paralinguistic editing (swap accents, inject emotion), and acoustic editing (remove noise, replace a single spoken word)
- Zero-shot voice cloning without reference transcripts, plus full TTS, vocal separation, and studio-grade cleanup
- Massive training scale: 3.03 billion instruction-audio instances and 1.95 million hours of effective supervision; reported 2.65% error rate
- Architecture: multimodal LLM for semantic conditioning + audio VAE trained on speech, general audio, and music + hybrid rectified-flow Transformer with dual-stream MMDiT blocks
- AuK-Flash: distilled via trajectory-level consistency init and task-routing decoupled DMD, reaching near-teacher quality in 4 steps without CFG
Aimed at dubbing, podcasting, and game audio. Uses Qwen3-Omni/ASR/ForcedAligner.
More from Multimodal
- Magnific teases October 28 release: 'We're making it move' — cuenca · 2026-10-27
- Midjourney --sref demo: retro Americana Batman art series — gizakdag · 2026-10-27
- AI video is a new medium, not cinema or animation — and it's perfect for depicting dreams — Kyrannio · 2026-10-27
- Dad turns his toddler's 31-year-old quote into a song via SUNO kids' choir — NathanpmYoung · 2026-10-27
- Band's local-AI music video hits 200k views: 900 generations kept 64, 3-second clips beat face drift — Hwang0500 · 2026-10-27
- Adobe Firefly video editor walkthrough: one-sentence Quick Cut turns 32s into 20s — LudovicCreator · 2026-10-26