Tencent has made the AuK foundation model for speech generation and editing available as open source, including code and weights on Hugging Face and ModelScope. The release includes AuK-Flash, a distilled variant designed for fast inference with only 4 steps.

  • AuK is a 1.5B model trained on millions of hours of audio data.
  • It supports zero-shot TTS, content and acoustic editing, paralinguistic editing, speech enhancement, and source separation via natural-language instructions.
  • The repository provides official weights for both the base model and the fast AuK-Flash variant.
  • Installation and usage examples are provided for Hugging Face and ModelScope platforms.

The release allows developers to access a unified interface for multiple speech tasks through the MIT license.