Tencent releases AuK: one open 1.5B model for speech synthesis and editing
In one sentence On September 9 Tencent Hunyuan open-sourced AuK under MIT: a 1.5-billion-parameter model that combines zero-shot TTS, voice cloning, speech editing, denoising and source separation, with a distilled 4-step AuK-Flash variant.
Tencent has published AuK for free, an AI model for speech audio that does on its own the job of many separate tools.
It can read text in a voice taken from a short sample, or in a voice described in words. It can edit an existing recording by changing some words, pitch, speed, emotion or accent. And it can clean noise out of audio or separate a voice from music.
Code and weights are downloadable under the MIT license, so they can be used in commercial projects too. There is also a faster version, AuK-Flash, built to generate in just a few steps.
Companies
Tencent
Tools
AuK, AuK-Flash, ComfyUI, SGLang-Omni, vLLM-Omni
Tags
Sources