Remove Background Noise from MP3

June 11, 2005 6:40 PM

I have tapes of interviews that I'm paying to have transcribed. Some interviews were done in public places and have background noise. I'll be charged more for those. Unless I can somehow get the background noise off the MP3s....

So my interviews are on microcasette. I've hooked my transcriber up to my laptop and I'm turning them into MP3s to send off to be transcribed. I'm using ReplayMusic to record the MP3s and MP3Surgeon to cut out the parts of the interview that aren't supposed to be transcribed.

One set of interviews were done in public places like coffee shops or malls. These can be loud as there is the sounds of other people talking, dishes clattering etc. Since this makes them hard to transcribe, they'll cost me more. I want to make them easier to transcribe by somehow turning down the volume on the background noise. Is this possible? How?

The transcriber has no equalizer, though it does have a tone switch (high or low). I don't have another microcasette player.

posted by duck to technology (7 comments total)

This would be forensic audio work. Though noise reduction is also used in restoration and general recording, they're not really what you're looking for. Adobe Audtion (formerly Cooledit), has a pretty slick noise reduction feature, as well some other useful filters. I've tried a couple of programs for restoring clipped files (distortion), but Audition is the only one I've used for noise. I tried out the demos of several Diamond Cut programs, but never touched the noise reduction features.

With audition, you try to find a segment with pure noise, and capture it as a "noise profile." The more consistent the noise, the cleaner the removal, but I've taken out background conversation, buzz, and traffic noises with great success.
posted by Jack Karaoke at 7:04 PM on June 11, 2005

Another option is Soundsoap, from Bias, a VST plug-in that works with any audio recording/editing app that supports plu-ins.

You are basically SOL, though. Microcassette is an awful medium -- I highly recommend switching to minidisc or better a solid-state recorder that is capable of 16-bit recording like the Edirol R-1. The original signal from your microcassette (I'm just gonna guess you also use a built in mic on the unit, which means you record the grinding of the motor, conveniently right in the heart of the human speech frequency range) is already going to be so noisy and have such a compressed frequency range that noise reduction and digitization (into a compressed format no less) is close to pointless.

Soundsoap is about $100, or in a pro version I think it's $500.
posted by realcountrymusic at 8:35 PM on June 11, 2005

Do you need the transcripts. Michael Frisch (_Shared Authority_, etc) is now heading up randforce and promoting indexing rather than transcribing. I sat through a demo of the technology and it makes sense to me - you might look into it an not have to shell out the cash for transcription (not that indexing won't take some work).
posted by jmgorman at 8:39 PM on June 11, 2005

The noise reduction technology works well...when it's a constant or repeating noise.

For computer algorithms to discern between which voices to keep/remove...well, that's where it gets messy. Very messy.

A noise gate will remove lower noises...and also part of the quieter part of the spectrum.

If you do any analysis - At least do it before you make the MP3 and trash much of the information (MP3 compresses the file, making it even harder to hear/remove the bad parts of the sound)
posted by filmgeek at 9:30 PM on June 11, 2005

During mp3 compression the noise you are trying to remove may distort the signal you are trying to recover, so remove as much of the noise as you can from the raw audio (a wav file or directly on the audio in). Lossy compression, like mp3, should be the very last step. filmgeek beat me to it, but I will offer my alternate explanation anyway... Sorry if it is too obvious...
posted by Chuckles at 9:37 PM on June 11, 2005

You know, I'm not sure the answer is technology -- I think the answer is a new transcriptionist. My last transcriptionist (who moonlighted for me from her day transcription job at a major news weekly) suffered through some horrible loud tapes, but would never have thought about charging me more. I always tip out my transcriptionists (at around 10, 15%) when I turn in a shitty tape -- and about half my tapes are really gnarly, in cafes, on the street, etc. -- but they have an hourly rate and that's what they get. I think your transcriptionists are fussy and you should get a new one.
posted by RJ Reynolds at 7:24 AM on June 12, 2005

Thanks for the advice everyone...I'm using the "noise profile" / "reduce noise" feature and it's working pretty well... I actually originally planned to transcribe these all myself, and I've done most of them, just the last 20 or so I'm farming out because I don't have time and I'm sick of it.

The tapes are about an hour long...I might try recording as a wav first, but geez that's a lot of HD space! Also, the files take about 10 minutes to open now and half an hour to "noise reduce" will making a wav increase that?

My transcriptionist is a company in India who charges $36/hour of tape. I'm not going to beat that with someone else, even if I do get charged for noise. If the charged hourly for transcription time, then the added charges would be built in, since a noisy tape would take longer to transcribe.
posted by duck at 8:21 AM on June 12, 2005