I recently came across old recording from the 1930's. Unfortunately not only is the quality low, it's a bunch of carpenters talking as they saw things and make all sorts of other noises.

I've tried:

  • Various filters
  • Audacity's native Noise Reducer

and yet no luck, I find that either what I'm doing is not working, or when I get the speech to sound louder than the tools the amplitude is too low to make out anything.

Is there any well defined technique for this?

  • $\begingroup$ Hi! Can you upload the audio clip here ? $\endgroup$ – Fat32 Mar 5 '19 at 21:59
  • $\begingroup$ Well @Fat32 I just ripped the audio from this video: youtu.be/wO2-ownJAYc?t=151 right now I'm just trying to analyze from 2:30 to 3:02. But if you'd like me to host the audio I'm gonna need a few more hours... $\endgroup$ – madprogramer Mar 6 '19 at 1:24
  • $\begingroup$ The audio in the video has most probably been restored heavily. There is a very low amount of broadband noise and other disturbances that are related to the age of the recording and the recording technology. Obviously, the audio bandwidth is rather low and the saw noises interfere with the speech. Unfortunately, there is no easy way to solve any of these two problems. $\endgroup$ – applesoup Mar 6 '19 at 9:02
  • $\begingroup$ Approaches for bandwidth extension for speech signals exist, but this is still ongoing research and I don't know any available solutions that are directly applicable for your case. To reduce the saw noises, perhaps iZotope RX 7's Spectral Repair can be of benefit, however, note that this kind of restoration is a manual process that may require a large amount of time. $\endgroup$ – applesoup Mar 6 '19 at 9:04
  • $\begingroup$ By the way, your attempts to use filters probably didn't bring you far because they simply reduce the intensity of certain frequencies. However, what is required in this case is time-variant filtering, i.e., filtering that changes its behavior over time: If speech is present without any interfering noise, no filtering should happen. If, at another time, saw noise is present, mostly the frequencies of the saw noise should be reduced, while the speech elements should remain untouched. Regarding your point "[...] the amplitude is too low": Why does increasing the volume not work? $\endgroup$ – applesoup Mar 6 '19 at 9:14

Your Answer

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

Browse other questions tagged or ask your own question.