Resemble Enhance is an open-source AI model from Resemble AI that denoises and enhances speech recordings. Given noisy, compressed, or low-quality audio, it removes background noise and restores clarity and fidelity, making voices sound like they were captured in much better conditions. The project has earned more than 2,400 stars on GitHub and was last actively updated in December 2024. Note that the listing link points to the GitHub repository, since the project’s demo page is a hosted trial space rather than a product page.
Real-world speech recordings are rarely clean: interviews happen in echoing rooms, calls come through cheap microphones, and archival audio carries hiss and distortion. Resemble Enhance attacks this with a generative enhancement pipeline that both suppresses noise and reconstructs damaged speech, rather than simply filtering frequencies the way traditional noise reduction does. The model can improve intelligibility and perceived quality of recordings for podcasts, voice overs, call archives, and speech datasets. It is distributed as a Python package with pretrained weights, so developers can run enhancement locally in their own pipelines.
Resemble Enhance is free and open source under the MIT license. Running it costs nothing beyond your own compute; Resemble AI’s separate commercial voice products are not required to use this model.
Resemble Enhance is for audio engineers, podcasters, and developers who need to rescue imperfect speech recordings or standardize audio quality at scale. It is equally useful to speech researchers preparing cleaner training and evaluation data.