Thanks very much for the all of the feedback. Here's a bit more information about the original material, and what we're trying to do with it. We are working with the audio-visual recordings of the court proceedings of the UN International Criminal Tribunal for Rwanda. We have produced WAV files of the original audio material (we also have video recordings which are going to be digitized as well). These are called the Unredacted Master files, and will not be altered in any way.
In order to make this material accessible to the public, it needs to be redacted. The primary reason for redacting the material is to protect the witnesses who have agreed to testify under the condition that their identity remains undisclosed to the public. This is crucial, because revealing the identity of a protected witness could potentially put him or her in a life-threatening situation. As a result of the redaction, we are producing WAV files (the Redacted Master material) as well as mp3s (the Redacted Browsing material), both of which can the be made accessible to the public. We expect that most users will be fine with using the mp3s -- the Redacted Master files are being created for professional use (ie news broadcasts, documentary film productions etc), as well as making sure that we have a preservation-quality copy of the redacted recordings.
One aspect of the redaction is to remove any words or phrases which the witness, lawyer, judge etc may have mentioned in court which could reveal the identity of the witness (or another individual whose identity must be protected). This usually relates to the name of a location, the witness's occupation, relationship to the accused or someone else, details of a crime committed etc. The audio-visual redaction team is responsible for reviewing the recordings, identifying that information and removing it from the recording. These content cuts are being handled in much the same way that Richard describes -- a "neutral tone" replaces the word or phrase in the redacted version, simply to indicate that a cut was made (we've settled on 1/2 second for the duration of the tone).
The other aspect of the redaction is distorting the voice of the protected witness to ensure that anyone listening to the recording cannot identify the witness if they happen to recognize his or her voice (given the small size of many of the communities in Rwanda, this is a very real possibility). The voice of the protected witness can be heard in the recordings, and we need to find a way to alter it to the extent that it cannot be recognized, but of course is still understandable.
We have ruled out the option of simply altering the pitch. For one thing, this would cause big problems with sync when we start redacting the video, and we want to employ the same methodology for the video that we use for the audio. Another issue is that rasing the pitch would be very annoying to our users. Lowering the pitch has its own issues: considering the extensive use of sexual violence as a weapon during the genocide in Rwanda and the considerable amount of witness testimony in which these acts are described, the women witnesses need to sound like women. It would be both disturbing to the users and insensitive to the victims if the women giving this kind of testimony sounded like men. Another reason that altering the pitch won't work is that it's a very simple matter to reverse an alteration in pitch, and the distortion must be irreversible.
Using more complex effects, or combinations of effects, is also problematic, because, no matter how complex it gets, it is still possible to reverse engineer and make the voice recognizable.
We have considered the possibility of using compression to "destroy" the quality to such an extent that the voice of the witness could not be recognizable. The big problem with this is applying it selectively. We only want to distort the voice of the protected witness, not the voices of the lawyers, judges, interpreters etc. After all of the work that is going into producing redacted versions of the audio material, we don't want the end result to sound "bad". Also, using compression is obviously not an option with the redacted WAV files. (Identifying the portions of the recordings in which a protected witness's voice can be heard is a part of the redaction process.)
One idea that has been suggested is severely limiting the frequency bandwidth. This is similar to the idea of using compression: making bad quality work for us by "throwing away" information so it cannot be retrieved. Sort of an extreme version of a telephone voice. This seems like it could address most of the issues we've identified: it can be applied selectively and it shouldn't be too disturbing to the listener. We are doing some experiments with this option, but would be very curious to hear if anyone has any experience using this type of vocal distortion. Does it sound like it might work? Any other suggestions?
Any assistance would be greatly appreciated
Judicial Records and Archives Unit,CMS
International Criminal Tribunal For Rwanda
AICC Bldng Box 6016 Arusha
--- On Sun, 4/4/10, Richard L. Hess <[log in to unmask]> wrote:
From: Richard L. Hess <[log in to unmask]>
Subject: Re: [ARSCLIST] Voice Distortion
To: [log in to unmask]
Date: Sunday, 4 April, 2010, 12:57
Thanks for the clarification. Again, to try and understand this, let me restate the question as I think I understand it:
In digitizing a collection of audio cassettes to WAV file, the cassettes contain some confidential information that has been redacted by adding distortion in Adobe Audition, and then MP3 files have been created for public access with the distorted information included.
I would not trust that distortion could not be reversed to the extent of making the redacted material unaccessible.
What I have done -- with the clients' blessing -- is cut out the redacted material and replace it with a short 1 kHz beep that fades up and fades down fairly quickly. This lets the listener know that something was redacted, it saves the listener time since the entire duration of the redacted material is not included, and it is impossible to reverse. The only flaw in this is if you want the listener to know the duration of the redacted material, then you'd have to place silence of the same duration in the place of the redacted material.
I presume that the archives are retaining both a redacted and non-redacted WAV file of the material, as, at some point, the restrictions might be removed.
At 07:11 AM 2010-04-04, you wrote:
> Hi Richard
> Thank you, but just to explain our position. We have digitized from analogue audio cassettes to wav. file for preservation and intend to give access as mp3. However, before we do that we would want to redact all the confidential information in our recordings. What i would like to know is that:
> 1. If we have distorted the audio recording or is there a way that the original distortion can be undistorted by using the same software(adobe audition in our case) if the audio file has been compressed from wav to mp3;2. If it can, what is the best method of voice distortion in adobe audition without affecting the file
> Thanks for your assistanceIshumael
> --- On Sun, 4/4/10, Richard L. Hess <[log in to unmask]> wrote:
> From: Richard L. Hess <[log in to unmask]>
> Subject: Re: [ARSCLIST] Voice Distortion
> To: [log in to unmask]
> Date: Sunday, 4 April, 2010, 3:45
> Hello, Ishumael,
> If I understand your question correctly, there was once a WAV file that was then encoded as an MP3 and the WAV file went away and now you wish to reconstruct it from the MP3. Is this a quick assessment of what you wish to do?
> If it is, I'm afraid that the MP3 coding process throws away "inaudible" information or "barely audible" information, and I doubt there is any way of getting that back.
> When you speak of "distortion" if that existed in the original, there are some "de-clip" algorithms that are marginally successful if the wind is blowing from the right direction in slightly removing some forms of clipping, but, all-in-all, the experience I've had with these have barely been worth the effort.
> It would be best to somehow re-generate the original WAV files, preferably at 48 ks/s 24 bits for voice or 96 ks/s 24 bits for music.
> At 04:58 PM 2010-04-03, you wrote:
> > Dear colleagues
> > I do hope I find you well.
> > I would like ask for advice on the following:
> > 1. Is it possible to undistort a recording that has been compressed from a wav file to mp3 and manage to listen to the original recording if one was to use the same software (adobe audition) to listen to the recording2.What is the best possible way of voice distortion in adobe audition.
> > RegardsIshumael
> Richard L. Hess email: [log in to unmask]
> Aurora, Ontario, Canada (905) 713 6733 1-877-TAPE-FIX
> Detailed contact information: http://www.richardhess.com/tape/contact.htm
> Quality tape transfers -- even from hard-to-play tapes.
Richard L. Hess email: [log in to unmask]
Aurora, Ontario, Canada (905) 713 6733 1-877-TAPE-FIX
Detailed contact information: http://www.richardhess.com/tape/contact.htm
Quality tape transfers -- even from hard-to-play tapes.