I wonder if it's possible to 'clean up' an audio recording of a lecture from a smartphone, using some type of AI system?

  • 1
    $\begingroup$ Sure, but that's not really AI. You're looking at signal processing. AI would be if you looked at what the meaning of the recording; background noise doesn't really fall into that. $\endgroup$ Dec 12 '18 at 17:36
  • $\begingroup$ @OliverMason but aren't people starting to apply ML to these types of problems? (I seem to recall reading something about it in regard to video recently...) $\endgroup$
    – DukeZhou
    Dec 12 '18 at 22:17
  • $\begingroup$ @DukeZhou But what is there to learn? Noise is noise... You would just apply a filter to remove the noise frequencies which you can get from a statistical analysis of the data. I guess it's my general gripe with "AI" as a buzz word that is applied to everything under the sun. $\endgroup$ Dec 13 '18 at 9:15
  • $\begingroup$ @OliverMason noise is not "just noise". Even removing other people talk in background is a very non-trivial task, that can not be solved using regular signal processing methods. $\endgroup$
    – igrinis
    Dec 13 '18 at 13:21
  • $\begingroup$ @OliverMason I think that teaching an AI to remove any background noise and keep only vocal sound (so perfect for a lecture recording) would be very helpful for a lot of people. Since a lot of my recordings tend to be either quiet or have a lot of noise, while the only thing I want is the speakers voice $\endgroup$ Dec 13 '18 at 16:11

Yes, it is possible. Usually the noise reduction is done using regular signal processing methods, such as spectral subtraction due to demand for low latency. But of course, modern methods of deep learning is applicable to this problem. For example variational autoencoder is the first that come to my mind, you can check this project. As another example you can look at this project , which is based on WaveNet.

  • $\begingroup$ Thanks for the links, I'll check those out .Think it's what I'm looking for, yeah. I don't really need realtime processing. I just have a bunch of audio files from lecture recordings that I'd like to "extract the speaker's voice" from as much as possible. So hopefully I can do that with this. I'll mark your answer as the best answer when I've tried it out $\endgroup$ Dec 13 '18 at 16:15

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.