Cleaning a phone conversation

[lasso]

I've benn approached by a guy, that has recorded a telephone conversation. In it, you can clearly hear him talking but the other person is only recorded very faintly through the cell phones own microphone. THe largest problem is, that there is music running in the background which almost completely masks what the other person says. And since the other persons voice is only recorded through the cells own mic, the music is in the same frequency range as the speak. I've triede numerous noise removal filters but to nu surprise the results aren't satisfactory.

I've asked the client to find the exact music that was played and both send it to me raw but also have him record it through his cell. My hope is that I can mess the phase and somehow minimize the music in the file.

Do you have any other clever suggestions? I really wanna help this guy - it's xmas afterall.

[huejahfink]

Sounds tricky... do you have a clip to share?

[Slyderhodge]

I had a job just like this a few months ago. First off I duplicated the track and edited it down so that only one person's voice was on each track, this makes leveling and cleaning each voice much easier. Also if you can't eq the entire voice because of the background noise, try concentrating on the syllabence of the voice. The one I worked on the phone voice was rather high pitched so I had a very thight Q boosted at 1.7k (somewhere in that area). Also using a HPF and a LPF so that your only dealing with the the voice helps too.

Not sure if this would help

[Richard Black]

See if you can beg a favour off someone (most likely in sound-for-film) who has one of the CEDAR DNS-series units. They're specifically designed for recovering dialogue almost lost in noise.

[imaginaryday]

Originally Posted by lasso

I've benn approached by a guy, that has recorded a telephone conversation. In it, you can clearly hear him talking but the other person is only recorded very faintly through the cell phones own microphone. THe largest problem is, that there is music running in the background which almost completely masks what the other person says. And since the other persons voice is only recorded through the cells own mic, the music is in the same frequency range as the speak. I've triede numerous noise removal filters but to nu surprise the results aren't satisfactory.

I've asked the client to find the exact music that was played and both send it to me raw but also have him record it through his cell. My hope is that I can mess the phase and somehow minimize the music in the file.

Do you have any other clever suggestions? I really wanna help this guy - it's xmas afterall.

download a trial version of Auidonamix and give it a try.

[LirvA]

This thread reminds me of a deal I was recently watching about Robert Kennedy's assassination. They got ahold of a tape that was the only known audio recording of the actual gunshots, and an audio forensic scientist analyzed it and came to the conclusion that there were 13 shots fired, and two different types of pistols.

It's a pretty interesting audio related field imo.


fwiw here's one of the vids
http://www.youtube.com/watch?v=iwvO0u8MseA

[synthoid]

I've had decent luck with Soundtrack Pro for this kind of thing (I cleaned up a conversation recorded on a cell phone with it). In addition to basic noise reduction stuff that is not bad, there is a spectral editor that can be very handy for doing radical surgery on something like background music. If you have enough time to throw at the problem, you can work section by section applying different kinds of spectral cuts to eliminate that stuff.

Another thing that you might try, on a lark, might be melodyne's polyphonic mode. If the background music happens to be isolated by that program as a separate melodic line, then you might be able to pull its volume down while leaving the voice alone. It depends entirely though on what Melodyne gives you when it analyzes the file; very much hit or miss.

-synthoid

[sidamosae.be]

Originally Posted by LirvA

and an audio forensic scientist analyzed it and came to the conclusion that there were 13 shots fired ..
YouTube - RFK assassination acoustic evidence on CBS

By looking at the waveform?

So that's what the loudness war is about, every 'peak of high amplitude' in your recording can be identified as a gunshot ...

If you watch closely you can see someone threw 3 grenades too, by the look of the amp envelope ...

[LirvA]

I don't know what all he analyzed, I'm a layman. But a few other people have come to the same conclusion that there were more than 8 shots.

[blusound]

Originally Posted by Richard Black

See if you can beg a favour off someone (most likely in sound-for-film) who has one of the CEDAR DNS-series units. They're specifically designed for recovering dialogue almost lost in noise.

Hi Richard,
CEDAR is the top company for forensic noise removal, but the problem is the music in the background and there in no music removal filter implemented until now!
The traditional approach to this kind of problem is to try to find out a comb filter with the use of a delay and more EQ to pull out some speaking, but no guaranteed results.
Synthoid suggestion with melodyne polyphonic mode could helps!
Best regards
Leo

[DrDeltaM]

Originally Posted by blusound

Hi Richard,
CEDAR is the top company for forensic noise removal, but the problem is the music in the background and there in no music removal filter implemented until now!
The traditional approach to this kind of problem is to try to find out a comb filter with the use of a delay and more EQ to pull out some speaking, but no guaranteed results.
Synthoid suggestion with melodyne polyphonic mode could helps!
Best regards
Leo

Actually, Cedar Cambridge does have a forensic 'music removal filter'. The X-Channel Adaptive filters can remove the music (distorted by the broadcast, acoustics, speaker system, recording system, ...) if you give it a reference track (the song in question from another source like a CD).
You don't even have to align the reference track perfectly, Cedar will do the fine aligning in realtime for you.

[blusound]

Originally Posted by DrDeltaM

Actually, Cedar Cambridge does have a forensic 'music removal filter'. The X-Channel Adaptive filters can remove the music (distorted by the broadcast, acoustics, speaker system, recording system, ...) if you give it a reference track (the song in question from another source like a CD).
You don't even have to align the reference track perfectly, Cedar will do the fine aligning in realtime for you.

I'm speechless!

[nemo]

I thought about finding the original music track (if you can recognize it) and putting it to the parallel track in the reverse phase.

[DrDeltaM]

Originally Posted by nemo

I thought about finding the original music track (if you can recognize it) and putting it to the parallel track in the reverse phase.

Simple anti-phase isn't enough usually, especially as the track is 'deformed' a lot by acoustics, speaker system, recording system, ... The Cedar filter can 'adapt' to these sonic changes to achieve a much better result then simple anti-phase.

[Richard Black]

I thought about finding the original music track (if you can recognize it) and putting it to the parallel track in the reverse phase.

Yes, that sounds neat until you realise you need to align the two tracks to a few microseconds over the entire duration, and then ensure that the frequency response and associated phase shift of both are identical to tight limits. When you realise how long that could take you (er, think lifetimes, frankly) you'll see what a bargain a CEDAR system is!