Audio forensics

[linusmay]

This company claim to be able to lift and isolate specific audio frequencies and vocal artifacts from poor audio recordings.

http://datarecoverysweden.se/audio/

Has anyone ever used them?

I have some security camera files, were an argument outside our office turned into a fight against one of our employees.
The audio is not very clear and our lawyer wants the audio cleaned up, so we can identify who said what to whom.

(The other person is saying that our employee provoked him by what was said).

The conversation in the cam footage is almost inaudible because of traffic and background noise, but you can certainly see there is something bad being said between these 2 people.

So we are hoping there must be enough digital information of the conversation that can be enhanced, once all the other sound is taken away.
we want to see who is telling the truth.

Please let me know if anyone has used this service, or if there are other ways of cleaning up audio files.

linus
linusmay @ null.net

[Hugh Robjohns]

Sorry, I have no experience of that specific company.

An alternative that I can recommend, though, is CEDAR's bureau service.

https://www.cedar-audio.com/bureau/bureau.shtml

CEDAR is the world's leading designer of digital audio restoration processing tools, and one side of their business specialises in forensic applications. If anyone can extract something usable and provide an acceptable legal audit-trail, it would be the CEDAR people.

H

[Eddy Deegan]

.. or if there are other ways of cleaning up audio files.

You may not need to 'clean it up' as such, rather just to make any dialogue comprehensible.

First, make a safe copy of the audio as it currently is, so that you can readily go back to the original if required.

On the copy you've made, you can experiment freely. If you load it into something like Audacity (which is free software) and then select the audio track by clicking on it, you will find a number of sound shaping tools in the 'Effect' menu.

Within there are a number of EQ and filter options. You may find that experimenting with the EQ and/or a notch filter you can increase the frequencies around the spoken vocal range, or reduce the frequencies asssociated with traffic noise enough to make the words audible.

The reason for keeping the original is so that in the event it's needed, someone else (possibly a specialist of some kind) can work on it independently without anyone being able to accuse you of compromising the integrity of the process.

Depending on how obscure the vocal is in the original recording your results may vary but even with no experience, as long as you're working on the copy of the original, then it seems me to be worth a go.

There is an article on improving speech using EQ here: https://www.behindthemixer.com/how-eq-s ... igibility/

According to them (and this looks about right to me):

Our speaking voice has three frequency ranges that need to be understood;
1. Fundamentals. The fundamental frequencies of speech occur roughly between 85Hz and 250Hz.
2. Vowels. Vowels sounds contain the maximum energy and power of the voice, occurring between 350Hz and 2KHz.
3. Consonants. Consonants occur between 1.5KHz and 4KHz. They contain little energy but are essential to intelligibility.

Your mileage may vary, but that should give you some guidance on where to start with the EQ should you try the exercise.

Getting it done professionally is the best option, but may not be needed if some tweaking on your part makes it clear enough to use and the process you used to do it can be repeated on a fresh copy of the original in front of a legal person if required.

As I said above though, work on a copy of the audio, not the original!

[James Perrett]

I would echo Hugh's recommendation to contact Cedar's forensics bureau. There have been big advances in extracting different types of audio from background sounds over the last few years and I don't think free software is going to come anywhere close - especially in the hands of someone with no previous experience of this sort of work.

[forumuser840717]

I'm with James and Hugh on this.

You need to be able to prove the authenticity of the audio (and that includes the original source - not just the subsequent copies and any processed files) and strictly speaking that means audit trails, possibly requiring, but certainly available for, independent verification, preferably done by people legally approved to do the work, etc.. And as the other party will have the right to query the audio and any work done on it you'll need to make sure you closely follow whatever procedures your legal system likes to use.

In this case it sounds like it might be more tricky than a straight 'make this intelligible' job as it sounds like voice differentiation is required to be sure of who is saying what. Assuming that the video is clear enough, you might also want to try backing up any audio by the services of a court approved expert lip reader (if your local legal system permits such evidence) to help clarify who is saying what.

I used to be a registered expert witness for forensic audio work and court presentations and, if you're planning to use the audio as evidence in legal procedings, definitely do not play around with it yourself using some random free (or paid for) software you find online. Consult (or get your legal team to do it) whatever registers exist in your jurisdiction for court approved practitioners. It may cost more than a DIY job but will be so much less messy in the long run.

[Eddy Deegan]

... and in light of the above, I'm certainly not going to disagree with them!

[Wonks]

Just a suggestion but, taking into account all the above, it may still be worth having a go on a copy with some quick and simple tools, just to see if you can make out the dialogue.

If you can determine what the true situation was, by using some low and high pass filtering and selective EQing, then you might work out whether the employee was provoked or did the provoking and then whether you want to proceed with getting the audio processed at a professional forensic level to take the case forward or not.

Of course it may not be possible to do this. However, if it works, it might save some money if you the decide not to proceed further.

[The Red Bladder]

https://www.izotope.com/en/learn/matt-boudreau.html

[Tim Gillett]

Much of this work is experience in listening to the available audio, and quickly getting a sense of what improvement in intelligibility is likely. Reputable operators may offer to listen to the file carefully and offer their opinion on the likely improvement they could effect in voice intelligibility. Most people's views as to what is possible has been skewed by CSI type TV dramas.

[linusmay]

Thank you all, so very much for your suggestions.

I have some good information now on how best to proceed.
I will update this, periodically as the workflow progresses.

Great input.

kr,
linus

[ken long]

Nice thread. Great advice from forumuser840717.

I'm assuming you've disclosed the recording with the other side's legal team. I'm sure your lawyer will have done just that.