I have an interview recorded using a Zoom H1n audio recorder and and two DSLR internal mics. Is there a way that I can use all three tracks in some sort of noise reduction process that looks for the loudest common sound in all the recordings (the interviewee speaking)? Something similar to adaptive noise reduction using multiple microphones that is carried out by modern phones?
6
-
Though I don't know how you would achieve it, i'd say only if none of the mics changed position over time, otherwise you'd be chasing phase for the entire track too.– TetsujinCommented Apr 8, 2019 at 16:07
-
Yes all mics were fixed in position. Is there were some kind of Layering system in audio similar to Photoshop where the Blend Modes such as Subtraction, Difference, Overlay, Screening, etc. can be used? That would probably be what I think I'm looking for....– salmonlawyerCommented Apr 8, 2019 at 16:10
-
My recommendation is that if you need to capture audio correctly during an interview for a case, there are lots of professional sound recordists available with the right skills and the right equipment. It feels a bit like a sound recordist in a legal forum trying to get free advice on a lawsuit they are filing.– MarkCommented Apr 9, 2019 at 3:24
-
research cross-correlation functions. This is an audio/digital signal processing function which may be able to assist.– MarkCommented Aug 7, 2019 at 2:51
-
1@MS-SPO I have honestly forgotten what this was for, it was most likely a school project at the time. Anyway I have learnt a lot more about audio and processing in the intervening years than I knew then, and I would probably have just used a noise gate on the H1n's recording as the first stage and added some basic noise reduction on top of that. It would have been enough for what I was trying to do.– salmonlawyerCommented Apr 21, 2023 at 7:28
|
Show 1 more comment
1 Answer
There probably is not much of a point trying to combine the sources. I'd use just the H1n sound while using the DSLR sound as reference for synchronisation with the video (I assume you recorded video on the DSLR). If you have equal quality in all sources, Wiener filtering would give you about 3dB gain for every doubling of the relevant sources but you don't have equal quality. If you work with non-linear filtering like spectral subtraction, you can recover significant amount of recognizable speech at the cost of a weird noise structure ("musical noise"). If you hand-placed a H1n in an interview environment, I would be surprised if the gains from such a procedure would offset the artifacts they buy you.