You are here

Best approach for forensics audio?

Page 1 of 1

Best approach for forensics audio?

PostPosted: Wed Sep 04, 2019 2:47 pm
by Nickjoe
I have a copy of Izotope RX7 and I'm trying to figure out the best approach for a file.
Is best to amplify the recording at a lower volume as much as possible or amplify it and then work on it with the computer and speaker turned up to 7 or so?

Re: Best approach for forensics audio?

PostPosted: Wed Sep 04, 2019 3:25 pm
by Hugh Robjohns
It depends to some extent on what you're trying to do to the file, but if the source file is very quiet I'd use the Normalise module in RX to bring the peak level up to -6dBFS or something like that -- high enough to make it easily audible, but still leaving some headroom for any processing you need to do.

H

Re: Best approach for forensics audio?

PostPosted: Wed Sep 04, 2019 9:33 pm
by Nickjoe
I think I'm on the right path then. So far I used Normalise all the way to zero.
Should any noise reduction be used before or after amping it up?
I notice that Rx tends to add a little gain if you use repair assistant.
I've also tried Proximity Eq which really gives it yet another boost. There is hiss in the file, but if I use spectral to get rid of it, it does become quite a bit phasery.

Re: Best approach for forensics audio?

PostPosted: Wed Sep 04, 2019 10:36 pm
by Wonks
If you've gone to 0dBFS and then processed it, then you've almost certainly distorted any exported files. Whilst working in the DAW, you'll be working on floating point files with loads of overhead, but once you create a 24 bit wav file (or similar), you'll need to make sure that it was exported below 0dBFS, and hopefully below -1dB on a true peak meter reading.

It really is best to keep all the processing at or below that sort of figure. Reduce input and/or output gains of the processes if necessary.

I trust you have the original files safely saved? If so, I'd start again, normalising to -6dBFS and trying to keep the results of each processing step around that sort of figure.

Re: Best approach for forensics audio?

PostPosted: Wed Sep 04, 2019 11:19 pm
by James Perrett
Nickjoe wrote:here is hiss in the file, but if I use spectral to get rid of it, it does become quite a bit phasery.

This is where multiple passes of slight noise reduction work better than a single pass with lots of noise reduction.

Re: Best approach for forensics audio?

PostPosted: Thu Sep 05, 2019 3:01 am
by Tim Gillett
Nickjoe wrote:I think I'm on the right path then. So far I used Normalise all the way to zero.
Should any noise reduction be used before or after amping it up?
I notice that Rx tends to add a little gain if you use repair assistant.
I've also tried Proximity Eq which really gives it yet another boost. There is hiss in the file, but if I use spectral to get rid of it, it does become quite a bit phasery.

Forensic audio is a much misunderstood term, not least due to TV shows like CSI. It's miles away from "I used Izotope Voice Denoiser and got rid of all the background noise"... Typically a forensic audio person will be trying to objectively increase the intelligibility of covert voice recordings, measured in intelligible words, and specifically the number of extra words made intelligible after processing.

https://www.soundonsound.com/techniques ... ngineering

YT is littered with examples of voice recordings before and after so called processing, but of the many examples I've checked, almost every voice was already fully intelligible before processing.

Specifically, what are you attempting to achieve here ?

Re: Best approach for forensics audio?

PostPosted: Thu Sep 05, 2019 4:47 pm
by Nickjoe
The problem is I can make out what is being said in certain parts on the original, despite the low volume. I think years of copying guitar parts off records probably help in that regard.


However, I have to get it a fair amount louder for everyone else to really start hearing the voices well. The problem was they were having some problems understanding it.
I was using Diamond cut forensics at that point, mainly just trying to amplify the vocal sounds and use a bit of eq. So the voices were there, but for everyone else it wasn't intelligible enough. I guess inexperience on my part.

But now with Izotope I'm getting much better clarity and it is getting much more understandable.
So it's low volume, but, with some quite understandable parts.
When amplified, there is a lot of hiss and some room reverb.

Some parts of the track are better than others. So we can't get the whole thing perfectly clear, but there is more than enough.
It's a bit like this maybe, but more inconsistent in volume, with more noise turned up.
So not quite as good.


https://www.youtube.com/watch?v=Xq_3kE6Al2Y

Re: Best approach for forensics audio?

PostPosted: Fri Sep 06, 2019 1:28 pm
by James Perrett
That example doesn't sound genuine - the before sample sounds far too muffled to be a genuine recording. There will always be more high frequency hiss on a genuine untouched recording.

Re: Best approach for forensics audio?

PostPosted: Fri Sep 06, 2019 2:41 pm
by Tim Gillett
That example and others from that channel seem mere advertising. No attempt is made to teach how they got from before to after, ie: what tools and what techniques were used.

Often the best way to increase intelligibility in a sample of that sort is EQ. Noise is only a problem if it reduces intelligibility. In the example 3 I basically rolled off a lot of lows and greatly boosted speech highs and achieved roughly the same intelligibility as they did.

There's nothing new in this. Operators since the early part of last century did the same sort of EQ to maximise speech intelligibility over low quality long distance phone lines and radio.

Example 1 from the same channel sounds faked to me. The bandwidth of the noise far outstrips that of the speech, suggesting the noise has been artificially mixed in with the audio later, and is not part of the original recording. Even so, they could have told us the noise is mains hum with lots of harmonics, and that the standard tool to reduce it is Dehum. But I guess they aren't about to tell us their secrets...

With experience (and tbh natural ability) we get a feel for which tool will help even before we pull it up, or in many cases with appallingly poor audio, we can know pretty much just by listening when no tool will get us there. Ref the SOS article I cited earlier by James Zjalic.

Re: Best approach for forensics audio?

PostPosted: Fri Sep 06, 2019 8:13 pm
by Nickjoe
That article does lay out the basic problem in that some people want the recording to sound pristine. I can tell what is being said, but everyone else seems to want a noise free type of solution, whereas a little bit of noise reduction and eq seems to work. In any event I'm going to see what I can do with some of the good advice here.
Would a screen shot of a section help?

Re: Best approach for forensics audio?

PostPosted: Sun Sep 08, 2019 3:57 am
by Tim Gillett
If the problem is big contrasts in the voice volumes, and the recording is to be played before an audience, I wouldnt see a problem in bringing up the quiet speech sections, or bringing down the loud sections if it helped people understand what's being said. Of course it would need to be made clear that that processing had been made.

A screen shot might convey an idea of the dynamics involved but of course the real audio is always the best, though completely understand the privacy issue.

Re: Best approach for forensics audio?

PostPosted: Sun Sep 08, 2019 11:38 am
by Mike Stranks
I know I've banged this particular drum several times before, but I find using a software Broadcast Processor is a great help in smoothing dynamic range in recordings - ie boosting quiet and pulling-back loud.

You don't have to spend anything1 The one I use is free, is well-respected and came bundled with my DAW.

http://www.vst4free.com/free_vst.php?plugin=Broadcast&id=548

Re: Best approach for forensics audio?

PostPosted: Sun Sep 08, 2019 12:04 pm
by John Egan
Mike Stranks wrote:I know I've banged this particular drum several times before, but I find using a software Broadcast Processor is a great help in smoothing dynamic range in recordings - ie boosting quiet and pulling-back loud.

You don't have to spend anything1 The one I use is free, is well-respected and came bundled with my DAW.

http://www.vst4free.com/free_vst.php?plugin=Broadcast&id=548

Mike,
That's a useful piece of software. Is there an equivalent 64 bit product that you know about ?
Regards, John

Re: Best approach for forensics audio?

PostPosted: Sun Sep 08, 2019 2:05 pm
by Mike Stranks
John Egan wrote:
Mike Stranks wrote:I know I've banged this particular drum several times before, but I find using a software Broadcast Processor is a great help in smoothing dynamic range in recordings - ie boosting quiet and pulling-back loud.

You don't have to spend anything1 The one I use is free, is well-respected and came bundled with my DAW.

http://www.vst4free.com/free_vst.php?plugin=Broadcast&id=548

Mike,
That's a useful piece of software. Is there an equivalent 64 bit product that you know about ?
Regards, John

Not that I know of John. I use this as a VST in my various bits of software on 64-bit Win 10 with no issues.

Re: Best approach for forensics audio?

PostPosted: Sun Sep 08, 2019 3:05 pm
by John Egan
Thanks Mike. That's useful to know. I'm also using 64 bit W10.
Regards, John