4
$\begingroup$

I have a recorded interview with two speakers, in which one voice's gain was much higher than the other (by three orders of magnitude). Increasing the gain does not help, because speaker 2's voice is so quiet that it is hitting the digitization limit (I believe). Audio quality is not important, but it would be very helpful to have an approximate technique so that the words may be intelligible.

If helpful, I can try to post an audio example of the specific distortion here. The cadence of the voice is there, but it's so mangled the words are unintelligible. Basic interpolation and filtering (both long shots) were not successful in recovering the voice.

Any suggestions for analysis techniques and/or software would be much appreciated.

$\endgroup$
  • $\begingroup$ Is there a gain difference because there was only one microphone and one of the speakers was too close to it? $\endgroup$ – ruoho ruotsi Aug 19 '15 at 17:38
  • $\begingroup$ Essentially, yes. $\endgroup$ – user59071 Aug 19 '15 at 21:54
1
$\begingroup$

I would suggest dynamic range compression as a first port of call. Set the threshold low and the ratio quite high. Then apply make-up gain to an appropriate level. This functionality is built into many free audio editors such as Audacity. Matlab or Python implementation is also reasonably simple.

$\endgroup$
  • $\begingroup$ With dynamic range compression, the vocal track is almost intelligible, with some recognizable worlds in the noise. I'm wondering now about additional steps to boot intelligibility. Thanks! $\endgroup$ – user59071 Aug 24 '15 at 19:52
1
$\begingroup$

You can also use Automatic Gain Control(AGC), for such kind of scenarios where there is huge difference in magnitude.I think this matlab script might be usefull - http://in.mathworks.com/matlabcentral/fileexchange/11202-automatic-gain-control

$\endgroup$
-1
$\begingroup$

I might try Audiophonic, a web-based podcast post-production service. It does exactly the type of processing you're wanting automatically. It's free for the first two hours a month of audio (as of this writing). It includes an "intelligent leveler" and "loudness normalization". If the volume difference is too extreme, a fully automated solution might choke. But it's worth a shot. Or you could do some of the tricks mentioned in other comments to get it in better shape and let Audiophonic do the rest.

$\endgroup$
  • $\begingroup$ May I ask why I got downvoted here? I don't have any connection to this company so it's not marketing, and it looks like this software could solve a thorny problem in a couple clicks. How is that bad? $\endgroup$ – user2348114 Feb 13 '17 at 5:27

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.