Hacker Newsnew | past | comments | ask | show | jobs | submitlogin
Show HN: free studio quality noise reduction for music (tape.it)
14 points by earthnail on Oct 18, 2023 | hide | past | favorite | 3 comments
Dear HN, this is Thomas from Tape It.

We're releasing our studio quality noise reduction software as a free web app today. You can try it at: https://tape.it/denoiser

Listening examples: https://www.youtube.com/watch?v=gRjxY7GTV6Y

We've been researching on this for over two years, and will present the paper next week at the Audio Engineering Society conference in NY. You can read about the research here: https://tape.it/research/denoiser

Short version: we combined a classic DSP denoiser with a neural network that controls it. Training this it was non-trivial, and we describe in the paper how we did it. Our research website also has a 2min video that summarises the paper.

The denoiser only removes stationary noise, like fan noise or amp hums. It works on all kinds of audio, not just speech, and we particularly focused on its applicability in music. Our main goal was to create a denoiser that always works and introduces minimal artefacts. To our knowledge, nobody has built such an automatic high-quality denoiser yet.

Please check out the video above, as well as the listening examples on our research page, where we also compare it against other software such as large deep learning models like Facebook's HDemucs, the very popular RNNoise, and non-automatic commercial software like iZotope RX.

We believe that this is a really useful piece of software. If you record audio, you can basically always run your raw recordings through our denoiser and assume that they'll sound better afterwards. Try it with your guitars, your vocals or field recordings.

Really looking forward to your feedback!



Very cool! Have you tried it on old 78rpm recordings, or is that a different type of noise than you’re aiming to remove?

[update] I've tried it on an old 78rpm recording and it successfully removed the hiss — but it was a bit too aggressive, so the resulting audio sounds muffled and lacking in presence/high end. I can completely understand that this particular use case might not be the target audience, though! I'll keep this link around for my own (modern) audio recordings.


We haven't tried it on old 78rpm recordings, but feel free to send one over to thomas@tape.it . The algorithm is trained on a surprisingly small amount of data by today's standards, due to the architecture we chose. There is a lot of room for improvement as we collect more examples, and it's something we definitely intend to do.


I did a quick test. It worked well.

Is there also a way to remove lip & mouth sounds?




Guidelines | FAQ | Lists | API | Security | Legal | Apply to YC | Contact

Search: