I'm looking for a source of very low-noise audio data for use in benchmarking various methods of random noise correction methods.

My intention is to have a very clean signal to which I can add known random noise such that I can test the efficacy of various methods of dealing with that noise.

It would need to be in some lossless format (preferably .wav).

Does anyone have suggestions?


  • $\begingroup$ What kind of audio material are you looking for? Are you looking for speech or music signals? In the easiest case, you can generate a low-noise audio signal by just using a sine. Otherwise: There are various sources with audio material licensed under the Creative Commons license (e.g. freesound.org or the Spoken Wikipedia project). If you let me know what exactly you are looking for I can write a more detailed answer. $\endgroup$
    – applesoup
    Apr 14, 2016 at 15:46
  • $\begingroup$ @applesoup I think music would be ideal, but speech would work well too. I've messed around with generating signals synthetically, but I'd prefer the signal to be very complex, more so than I can achieve by combing a few sine waves ect. The work I'm doing relates to errors from purely random noise (the stuff which gives you that white noise floor) so Ideally I need something with a low noise floor. I'll check out those links now - thanks for your time! $\endgroup$
    – CBowman
    Apr 14, 2016 at 16:47
  • 1
    $\begingroup$ How about text-to-speech engines? $\endgroup$
    – ThP
    Apr 14, 2016 at 17:11
  • $\begingroup$ What I like to do sometimes when I need really low-noise signals - be it speech or music - is to fetch some recording from the internet (maybe from the sites that I mentioned above) and run it through a denoising algorithm. I set the latter to a rather aggressive setting as maybe the signal distortion is less important than the reduction of the noise. This way, you can obtain very low-noise signals, however, that suffer from certain degradation of the desired (e.g. music, speech, etc.) signal. $\endgroup$
    – applesoup
    Apr 15, 2016 at 9:15

1 Answer 1


That is a simple question:

All sound that passes into analogue domain using digital audio converters (DACs) have a SNR inferior to about 113 at best, and more probably 96-100... everything recorded using microphones will have inherant noise, unless the recording has been copy pasted onto a silent track leaving intervals of silence.

So what you need to do to have a very clean signal, is to take a totally digital sound which was generated with i.e. sine waves and has not had reverb and echo effects that raise the noise in the silent time(s) of the recording, and digital synthesizers which return to 0 after the note envelope. You can go though a track that contains elements of silence in Audacity or a more easy to zoom editor, and check that you are having times in the track where the amplitude = 0.0; or program a fast test code to count that the track has amplitudes of 0.0 lasting for more than 500ms or something.

if you take any recording and vary the volume of the track, for example you take an AC-DC rock song, and apply an artificial envelope that fades the track to 0.00 regularly i.e. every couple of seconds, it also makes a test file with totally silent intervals.

you can also write a code that sais : if volume < than 0.05, volume =0; to artifically make any track you want totally silent at times.

find digital recordings on freesound and measure them for 0.00 intervals in between sounds, check digital music artists that boast about 192khz recording equipment for a particular album, get a feel of which artists use state of the art production equipment, check standard commercial tracks that are used to measure monitor speakers, there are many options.


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

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