Identify this type of audio file corruption?

[Seamonster]

I've encountered a bizarre sort of audio file corruption ? new to me ? which I'm hoping someone will be able to identify (link below), and perhaps to recommend a fix.

While I've been in the long process of building a new studio and a lot of my gear has been dismantled, I've taken work transcribing and editing interviews (for a memoir). In three of the many dozens of recordings made by the interviewer using a Sony ICD-UX71 handheld recorder, the audio suddenly becomes garbled. Typically this occurs 45 to 60 minutes into the file and continues to the end; the files are each from 1.5 to 3 hours in length.

I intuit that the distortion is caused by some sort of file corruption rather than anything electrical or mechanical happening during the recordings; that it could have been caused by something like the recorder's batteries running low during transfer of the file to a computer, a glitchy USB cable, or something of that sort. A web search reveals several other users experiencing similar "garblage" symptoms, but no solid conjecture on causes or remedies.

Here's a sample snippet. It's approx. 40 seconds long (6.7 MB); the voices are normal for the first 10 seconds, then the garbling kicks in. (There are other noises too ? the subjects are eating, or doing something like that.)

http://seamonstersounds.com/mp3s/FerbleClip.aif

I could attempt to isolate the signal from the noise using EQ or perhaps Izotope RX2, but I'm hoping there's a more surgical fix to be made on the files themselves ? perhaps using something like Soundhack? I don't have RX2, but surely it doesn't have a function that would fix the problem directly(?) (I do have the old Ionizer, so maybe I'd resurrect an OS9 boot partition to run it, before I'd spring for RX2.)

- K Hill

P.S. The first correct answer wins %&(*#! %& )%*($ #_@&%(#&....

[Seamonster]

Someone suggested the problem is "misaligned data" and/or "buffer underrun," which would seem to describe it exactly. So I half-wonder if there's some developer's tool with which I could identify the moment where the data goes out of alignment and surgically realign the data (analogous to zooming in tight on a waveform to re-draw a glitch).

But oh well, at least I can tell the client what she's up against, with some degree of precision.

[Gregg Juke]

Wow, never heard anything exactly like that before. Sounds like he turned into an underwater Cylon, there...

I guess the mis-aligned data thing is as good a guess as any.

GJ

[Seamonster]

Dunno what a Cylon is (except that it's something in Battlestar Galactica, which I've not seen), but I believe you must be exactly right.

I've downloaded several apps that do analysis-resynthesis (or something similar), though haven't yet tried playing with them. They include: Spear, Photosounder, Tapestrea, rt_lpc and Mammut. Can't say whether any of those will help much, but it should be a fun exploration/diversion.

[apropos of nothing]

I find that media players have differing ways of interpreting codecs. It might be worth a try to see if the audio is decipherable with other programs. Specifically, I think I'd try Media Player Classic, VLC, Reaper and potentially SUPER to see if anything was willing to read the file correctly.

[apropos of nothing]

Also, you should see if you can play the files off the recorder, itself, if you have access to it.

[Seamonster]

The fact that the few bad files play fine for the first hour or so then suddenly go bad beyond that point suggests that it's not a simple question of the codec.

Loading them back onto the original recorder would seem like a good idea, but the same file-goes-from-good-to-bad concept applies. And I don't know if the client even kept the original mp3's (what the recorder generated), but we did check her earliest safety copies, which were conversions to WAV format.

I think the problem is beyond any salvation via file trickery; I'll just have to try audio tools ? EQ, noise-reduction, and/or analysis-resynthesis ? to see how much I can separate the signal from the noise. (Not sure how soon I'll get to it, as I have other projects on the burner; the client views it as a lesson learned at this point and is in no hurry for me to make the impossible happen....)