Less-than-ideal audio into audio documentary.

Former user wrote on 6/14/2021, 1:31 AM

I have several digitized interviews from cassette tape that I need to get into shape for an audio documentary. Originally, I just needed the tapes to be good enough for transcription. Now there is interest in arranging the material for others to listen. However, the interview recordings have several challenges. In addition to tape hiss and most interviews conducted via phone, some of them also have tones sustained throughout (maybe from nearby electronics) or the recorder put in the middle of the room for all to speak (instead of directly at an interviewee). I was wondering if I could post segments from each tape that really has trouble and ask how would one approach the audio to get it into the best shape possible. What SF tools would you pick and how you would use said tools to elicit the best (within reason) audio?

Back to post

Comments

Former user wrote on 6/21/2021, 12:47 PM

@Rednroll Thank you for the posts and advice, however, this specific approach you advise does not work for much of the audio I have. I have tried this way you suggested several times over days and I keep getting that 'robotic' voice at the end. If I lower attenuation, I get too much noise. There is no 'compromise realm'. I have indeed provided the program with sufficient 'learn' time, but this is not enough. That is why I try out different audio tools and trials of programs. The piece I posted is one of the worst of the lot.

emmrecs wrote on 6/21/2021, 1:25 PM

@Former user

I keep getting that 'robotic' voice at the end. If I lower attenuation, I get too much noise

Which is why I previously suggested to you running a series of NR passes, each time taking off a little more of the noise. In practice, this is exactly how iZotope's RX works, it just manages to do it in "one pass". If, quite reasonably, you cannot afford the purchase cost of RX 8, then you have to "manually simulate" the same process.

Jeff

Former user wrote on 6/21/2021, 1:32 PM

@emmrecs I will think about this. Am not sure how much noise to take off at each pass. Perhaps target a specific sound each time...? Tape hiss for pass one, demon tone pass two, hum pass three...? Will I have any audio left after so many passes? Will look closer at SpectraLayers, am still familiarizing with the tools.

emmrecs wrote on 6/21/2021, 2:05 PM

@Former user

For each pass you would "take off" only a few dBs, perhaps as little as 3 or 4.

One thing I forgot to mention, assuming the software you are using is able to "learn" what is "noise", before each pass you would need to allow it to relearn the noise profile.

Jeff

Rednroll wrote on 6/21/2021, 2:15 PM

@Former user

I keep getting that 'robotic' voice at the end. If I lower attenuation, I get too much noise

Which is why I previously suggested to you running a series of NR passes, each time taking off a little more of the noise. In practice, this is exactly how iZotope's RX works, it just manages to do it in "one pass". If, quite reasonably, you cannot afford the purchase cost of RX 8, then you have to "manually simulate" the same process.

Jeff

The robotic voice is due to too much signal cancellation being applied at once, resulting in comb filtering of the signal you are intending to keep. It is a very similar FX as if you applied a "Phaser/Flanger" FX to the voice which are effects which use comb filtering to get that type of sound.

Lowering the amount of noise reduction being applied and performing multiple passes instead while relearning the still present noise on each pass as @emmrecs recommends is the best solution to achieve better results while minimizing the impact it has on the voice.

I'll also add, selecting an area which has only noise and no voice for the "learn" function is key also to obtaining the best results. The bigger that selection of only noise, the better.

Rednroll wrote on 6/21/2021, 2:20 PM

@Rednroll Thank you for the posts and advice, however, this specific approach you advise does not work for much of the audio I have. I have tried this way you suggested several times over days and I keep getting that 'robotic' voice at the end. If I lower attenuation, I get too much noise. There is no 'compromise realm'. I have indeed provided the program with sufficient 'learn' time, but this is not enough. That is why I try out different audio tools and trials of programs. The piece I posted is one of the worst of the lot.

Re-upload that poor audio recording but this time include some head and tail on it of just the noise by itself prior and after the voices start. If I get some time, I'll play around with it to see what can be done as far as improvements. I enjoy a good challenge, but need the essential elements to achieve the best results.

The good NR plugins allow you to listen to just the noise you are removing during playback. If you listen to that and hear any voice in it, then that means you're applying too much NR and the result will end up being comb filtering to the voice causing that robotic type of voice to be heard.