My girlfriend talks in her sleep in a language neither of us have heard before. It's a rare occurrence: Since we started recording our dreams in early December, we've caught just a handful of words. We're trying to capture as much as possible in hopes of understanding more about the language and hopefully one day decipher what she's saying.

The problem: Combing through 8 to 10 hours of audio each day is a painstaking process. I hung a decent mic from the ceiling and am using a Digidesign Mbox 2 Mini to capture the audio with Audacity. We've gotten real good at picking out patterns in the waveforms and skipping to the interesting bits, but we have to listen to hours of rustling, rolling over, etc, in order to find those precious few seconds of sleep talking. It's like looking for a needle in a haystack.

I'd love to speed this process up. Maybe some method for isolating vocals from the background noise? Or automatically skipping past anything under a specific db during playback (so we could play cards or do something else while listening)? Perhaps this community has other ideas?

Any thoughts or suggestions would be very much appreciated. Thank you!

2 Answers 2


Use a gate! The following process will let you isolate what you need.

First of all find the frequency of the voice, i suspect for a female voice an area between 250-800 should capture the main body of the voice.

Roll overs and sorting should be in a different spectrum.

Ater finding that area cut everything else and just bandpass that area. (you can even accentuate the frequency that responds best to the voice. )

Now take a gate that can listen to a frequency as a key (i know cubase's stock gate can do this also with a different Q) and set it so it opens only when she speaks.

After testing that it works every time and close to the way you want it, create a new channel that records the main channels output. (maybe google on how to route a channel to the other)

The result will be a buss with big silence and waves only containing the content the gate lets through.

Now that channel has very bad sound because of the bandpass & EQing but its content is not destined for listening to.

What the new buss will give you is the areas where the content is present. So after recording take the gate & eq out and listen to the areas where the 2nd channel has content!

Thats all, comment if you have questions!


Many handheld recorders have a feature that automatically starts a recording when a certain dB threshold is reached. That might save some time and file space. You'll have individual files of hopefully just whatever alien language your girlfriend is speaking!

I know the Zoom H4n has this feature. I'm sure cheaper ones do as well. Zoom calls the feature Auto-Record I believe.

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.