Need help with extreme noise removal/voice recovery

Audacity 2.0.6, Win7x64 SP1

I need help from someone more skilled than I am at recovery. I have an audio track of a recorded interview, but it seems that something may have come loose in the recorder, or the microphone been obstructed, or something of that nature. I now have a file of loud, constant white noise and very, very quiet distorted voices.

I’m not as good at editing as I once was, and I was probably never good enough to fix something like this. I would greatly appreciate anyone who would be able to recover the voice audio from this file. If this isn’t the right sort of place to ask for that, kindly direct me somewhere that is.

The total file is some 220MB and 90 minutes long. There are thee voices on it: an old male, an old female, and a young female. I think the sample I’m providing has at least the young female voice in it.

It was recorded with a cheap USB recording device; I don’t know the make or model. Other files from the device have been fine before and since this recording.

Like I said, any help would be greatly appreciated.

I think the sample I’m providing has at least the young female voice in it.

We can’t do forensics. If you’re in the “I think there’s a voice there” territory, you’re way beyond anything we can do.

You should start planning what you would do if you had no show.

The clip has very odd sounding voices. Did you try to rescue it already? We warn people to only post straight, uncorrected clips. If you messed with it already, then the clip is corrupted.

We also don’t do law enforcement.


This is the original audio. I suppose I was really just asking for a confirmation that it was impossible. Also I have no idea what you mean about law enforcement. Thanks for the input.

“I think my employees are stealing from me and I have the proof on this audio file that I can’t understand. Please help me put my employees behind bars.”


What was your show? Straight interview that went bad because of technical reasons? Our next concern is another failure if you try it again. It sounds a little like you were using a very nice microphone, but Audacity switched to a laptop built-in microphone instead. Could that be what happened?

If you lose connection to a USB microphone even for a second, Audacity will “forget” about it and go looking for something else. It will not come back by itself. Could that be what happened?


The clip you supplied is also primarily electronic noise like a very badly adjusted microphone (why I suggested an unintended microphone switch). Buried in the electronic garbage is something that might be a voice if it was louder.

I see no rescue. The noise reduction tools need a sample of noise without the voice in order to do anything. If we can’t tell if the voice is there or not…


Hah, no. Nothing like that. It was an interview conducted as a part of a friend’s doctoral thesis. She knew I used to do editing work and asked me to take a look.

And you misunderstand - I don’t mean a USB microphone, I mean an independent recording device. It couldn’t have switched to another mic because it wasn’t connected to anything when recording. It’s just a USB stick with a mic that spits out wav files.

Unfortunately there’s not really a chance of redoing the interview. She’s just going to have to reconstruct it from memory as best she can. It was never about having the audio itself, just about having the content that was discussed.

Like I said, I kind of knew it was a long shot. But It’s been five or ten years since I did any serious audio work, so I didn’t really know what state the recovery process was in. Figured I’d at least get a second opinion before telling her it was a lost cause.

It’s just a USB stick with a mic that spits out wav files.

I have one of those. Two, actually. I quietly turn it on when I’m expected to remember a long list of instructions. I nod my head reassuringly and then go somewhere to transcribe the sound file.

I’ve never had either one do that. Is it busted? If you hold it in front of your face and say “hello,” does it work? I know mine have two design problems. They are insanely susceptible to handling noises and they have very serious background hiss. I bought them to do interviews and that just didn’t work.

Were you the recordist? Someone with no live experience may turn it on and stuff it in their purse or run bag. Cover up the microphone holes and the sound is what you have.


By the way, there is something broken about the file. My Mac refuses to give me a “wav” icon on the desktop and the right sound channel has serious negative DC offset (overall too low - attached) after amplification.

So this is not a perfect example of sound recording in many different ways.

Screen Shot 2014-11-05 at 6.20.32 PM.png

It also has a click that repeats every 170 ms (dead on), and has a pronounced narrow frequency peak at about 11360 Hz.

It’s quite easy to substantially reduce the clicking, but the “pink” noise is just too loud compared to the alleged voice to get anything useful.
This is my attempt - it fails:

My USB stick records at 32000 (FM quality), not 44100. And it’s mono. It’s a close race between storage and battery life, but I think people have produced sound files for six straight hours. The stick knows to break up the production into individual WAV files.

It was sooo close. If only it didn’t fffffffff.


No, I wasn’t there. She just sent me the file, but I assure you she knows what she’s doing with it. Done plenty of these with it, and this is the only one gone bad. Tests afterward work fine too.

She mentioned that she had a hard time getting software to recognize it. Quicktime worked, but VLC and WMP didn’t. I can open it in VLC, though, so maybe she’s out of date on that. I’m not sure why you’d have an issue recognizing the sample clip - it was exported straight out of Audacity. The shifted right channel is like that in the original file too.

And yeah, that attempt is more or less what I ended up with, trying myself. The more I isolated the voice the clearer it is that there’s just no definition to it at all.

My best guess is freak hardware failure of some kind. It wasn’t a problem with the transfer (a second import gave the same garbage), so it must have happened during recording.

Paul-L’s de-clicker plug-in can get rid of the recurring click, but the hiss-noise is too loud compared the the voice signal to be comprehensible …

My best guess is freak hardware failure of some kind.

Moon Phase Error.
It’s last quarter Autumn Moon. Time for the sound to go in the toilet.

Actually, there’s a clue in one of the errors the file has. There’s no reason for the sound to have periodic clicking in it. USB Stick Recorders have no hard drives to share or other system processes to compete. That probably means the USB Stick was subjected to intense electronic interference during the performance. They have little or no shielding for interference.

It’s almost like the little recorder was placed too close to a powerful computer during the performance.


Interesting. I’ll have to ask where she was that might have had something like that.

And that second attempt there is really impressive. Useless, of course, but impressive.

We can turn the damage into different damage.

Generally if you can’t understand what is being said on the original no amount of signal-processing will make it comprehensible, [ the “Hollywood” forensic-audio is usually impossible science-fiction ].

Also digital-signal-processing adds artifacts which can sometimes sound like one or two words, which could make the processed recording inadmissible as evidence.

A note. Even in CSI or the movies, “evidence” like that is usually used as a means to get the real evidence, or in some scenarios, the perp breaks down and confesses to the crime making further research unnecessary and concluding the show in the required 42 minutes (television) or 93 minutes (movie).

"Wait, that’s not Alice’s car, that’s John’s car. We’d better go to John’s house and dig up the dead body from behind the garage.