MMAudio

MMAudio is a research-focused AI system for precise, text-driven audio editing and generation. It enables users to modify existing audio by describing changes in natural language, such as “remove the background chatter,” “make the speaker sound older,” or “add light rain in the background,” while preserving the original content and timing. The model supports multimodal conditioning, allowing edits to be guided by text prompts, reference audio, or a combination of both, which is particularly useful for style transfer, timbre matching, and consistent sound design across multiple clips.

MMAudio can perform localized edits, where only specific segments or attributes are changed, as well as more global transformations like adjusting ambience or overall acoustic characteristics. Key capabilities include robust content preservation, fine-grained control over what aspects of the audio are altered, and compatibility with a wide range of everyday audio scenarios such as speech, environmental sounds, and simple music.