How do I clean up a voice recording?

December 21, 2005 2:34 PM

How do I clean up a voice recording?

I have a WAV file of a friend of mine singing (with piano accompaniement). The recording isn't the best quality, so I would like to clean it up, and possibly enhance the vocals. There are a million sound editing apps but I have no idea which is best. Can you recommend some free applications, and/or ways that I could get the recording in better shape?

posted by blue_beetle to computers & internet (19 comments total)

I don't know of any free progs to do this with, but as someone who does this for a living I can tell you not to expect professional results without knowing what you're doing. The type of noise in the background will play a big part as far as your choice of noise removal, and if the recording is already mixed down into stereo (meaning that you don't have one recording of only voice, and one of only piano) you're going to have a tough time making the voice sound better without affecting the piano (which occupies much of the same frequency range as a singing voice).

Off the bat, I'd recommend simple EQ filtering. You can play back the WAV in winamp and change the EQ settings until it sounds good* and then take a program and record the output from the Windows Mixer option in the windows mixing panel.

All in all, you could fade in/fade out of songs, normalize the volumes to be similar across the entire recording, and not much else without having the luxury of experience and/or a good mixing room.

*EQing a mix properly depends a LOT on the speakers you use to listen with. If you make corrections for your speakers, then take the new mix and put it in your car, it may sound completely different.

So what type of enhancement and/or noise removal do you need? Are there obvious problems, or only fine-tuning needs?
posted by plexiwatt at 2:46 PM on December 21, 2005

For what values of 'clean up'? Are you looking to remove mic noise in general? If so, there are various methods that can analyse a section of audio that contains *only* unwanted noise, and remove it from the rest. Audition and the free Audacity have something like this built-in; there's also a separate plugin for SoundForge and the stand-alone DART (which is quite effective but has a god-awful interface).

Other kinds of more specific interference/distortion may or may not be cleanable using special plugins and/or judicious use of EQ and dynamics.

'Enhancing' the vocals when they're already mixed together with the piano is unlikely to be simple. There's always the cheap cop-out route of putting a load of dynamic range compression and reverb on everything I suppose. But nothing beats having a good clean dry-vocal track to begin with.
posted by BobInce at 2:52 PM on December 21, 2005

you can get rid of noise/hiss as BobInce mentions above. the more noise you try to remove, the more artifacting you get. if you try any of the programs out there, you'll be able to play around with the noise reduction and find a good medium. one such program is goldwave. not pretty, either, but i suppose it works! it's trialware.

if you want, you could also xfer the file to me, and i could give it a shot. no pro by any means, but i'll do it for free!
posted by herrdoktor at 2:58 PM on December 21, 2005

The (not cheap and not easy) Adobe Audition is the program of choice. http://www.adobe.com/products/audition/main.html

It may be possible to find a free copy of CoolEdit (the predecessor of Audition). Note that it's different from CoolEdit Pro, which can also be found cheap on eBay.

See generally http://www.delback.co.uk/lp-cdr.htm
posted by KRS at 3:06 PM on December 21, 2005

I've used audacity, which BobInce mentioned. It's free (as beer and speech) and multi-platform. The noise reduction filter is built in and pretty easy to use (you "train" it on silent parts of the recording, then apply the filter to the recording as a whole). It takes a little trial-and-error to find the parameters that work best (most noise reduction, fewest artifacts). It has worked well for me to remove computer fan whirr from home recordings but I'm no audiophile so my standards are easily met.
posted by TimeFactor at 3:16 PM on December 21, 2005

a bit more info: The recording was made on a cheap digital voice recorder, so it's a mono track, there's a bit of static and conversation in the backgroud. It sounds "tinney".
posted by blue_beetle at 3:22 PM on December 21, 2005

If you want to know what I'm working with here is the file. Don't worry, I'm not expecting any miracles.
posted by blue_beetle at 3:29 PM on December 21, 2005

I recently did something similar, and found I preferred the quality of the noise reduction filter in Acoustica, compared to Goldwave or Audacity. Not free, but has a fully-functional trial period. But you could try them all and see which you like. Then you can do EQ for the tinniness. I don't think you're going to get it sounding really clean.
posted by transient at 3:31 PM on December 21, 2005

Technical Specs: Bitrate: 32kps, 1 channel(mono), 8khz sample rate, IMA ADPCM encoding.
Audacity doesn't seem to open it properly (possibly the sample rate is too low?) but it plays back fine with WinAmp.
posted by blue_beetle at 3:36 PM on December 21, 2005

audacity (free) has a nice automatic noise cleaning function, ngwave has a good eq (afair) and costs ~30$, adobe audition has _lots_ of cleaning functions

definitely try eq'ing it in foobar too
posted by suni at 3:44 PM on December 21, 2005

oops, too late, sorry : )
posted by suni at 3:45 PM on December 21, 2005

how long is the sample supposed to be? i get around 59 seconds, then it starts doing this max headroom thing for another couple minutes.

anyway, here's the sample converted from ACM to PCM. and here's the sample after it's been cleaned up a bit.

what you'll want to do is convert the sample from 8-bit to 16- or 32-bits, then work from there. i went through it and found any clicks/pops/random noises by ear, then just kinda click/drag/deleted chunks of the .wav. sometimes it's easy to see where the source of the noise is. for example: if it's clipped, you'll see the waveform spike real big-like. cut that part out.

sometimes switching to a spectral display helps, too. this was done to easily find the low-pitched noises. i did a teeny-tiny bit of noise reduction, too.

if you need me to convert the whole sample, shoot me some email or IM me. i dunno why it doesn't open in audacity, tho. it should!
posted by herrdoktor at 4:22 PM on December 21, 2005

Ouch. 8k and clipped. I'd try what herrdoktor says, along with EQing with a little bump in the 2k range, cut <1 25, and cut>4k.

May as well keep your editing light, you may make it worse by removing what little "air" exists in the voice.

Good performance though!
posted by plexiwatt at 4:41 PM on December 21, 2005

I just listened to it, and it seems pretty horribly clipped (digital distortion), which is very difficult, if not impossible, to counteract. If re-recording is at all possible, I would suggest that. Watch those meters! I tend to let my peaks during level check not exceed -10db, which usually gives me plenty of headroom (but then, I tell my performers to play or sing a little louder during level check then they would usually - the heat of the moment tends to add a couple of dbs anyway when the tape is really rolling).

If you have the option at all I would suggest recording at 44.1KHz (or better, but I can't really ever see the point), and 16 or (preferably) 24 bits, 'raw' PCM/WAV.

If you really can't re-do it, I'd suggest some gentle EQ as others have suggested, and maybe some gentle compression with a very warm compressor. This may sound counter-intuitive, but you're doing damage control here anyway, and the warmth of the compression may help distract the listener from the horrible clipping going on. But even then I would consider this material next to unworkable.

However, the performance is brilliant - if you hadn't said it was your friend I would have thought it to be some pre-war Austrian operetta singer!
posted by goodnewsfortheinsane at 5:09 PM on December 21, 2005

I'm on the development team for Adobe Audition, and it will definitely help clean this recording up, although at such a low sample rate, I can't promise anything close to a miracle. You can get a 30-day free trial at adobe.com.

If you've got more recording than you posted, such as general noise and ambience before or after the recording, you can use the noise capture tools to grab a profile and remove/reduce that from the piece. I tried with the small gap in time between the end of the song and the applause, but it wasn't really enough.

If not, then I'd just run the basic Hiss Reduction plugin on it. Let it get analyze the noise floor, adjust it to about 1 dB, and apply. it might still sound a little "tin canny" so you'll then want to play with the EQ to enhance things.

I did a quick pass on it and put it online. (3 meg mp3)
posted by durin at 8:21 PM on December 21, 2005

I'm an Audacity developer. Audacity's noise removal will work for removing background noise (hiss, hum, static), but it won't repair the clipping.

PostFish (by Monty, the creator of the Ogg Vorbis codec) includes an impressive declipper. I don't know it well enough to say whether it would work on your particular recording. Unfortunately PostFish is still experimental, and not easy to install (Linux-only; no binaries). If you're desperate and want to try it, contact me (see my profile).
posted by mbrubeck at 10:00 PM on December 21, 2005

Just to add to my comment upthread, the -10dB guideline I use is based on the assumption of 24 bit recording; this gives you a wider dynamic range than 16 bits - at 16 bit I would probably sneak in another couple of dBs and peak at -6dB or so at level check. But then again, this gives you even less headroom - high frequency transients are *fast*, and they can peak many dBs over the 'limit' you thought you had set for yourself at level check.

The paradox rests with the fact that you want to maximize the peak level before clipping (to get maximum resolution / use the most bits available), while also raising the nominal recording level to the highest while not clipping also (for the same reason). An odd marriage, indeed.

If this doesn't work out I suggest adding a hardware analog compressor (or limiter) in the signal chain *before* converting to digital.
posted by goodnewsfortheinsane at 6:40 PM on January 6, 2006

...with the "guys"?
posted by growabrain at 1:53 PM on February 23, 2006

I pose a similar question, yet it is live concert footage from a casio 7.2 exilim , I was to close to the speakers and the audio is "blown" distorted.

Is there filtering to get rid of the junk but keep a decent recording.
posted by proph8 at 3:08 PM on June 21, 2006 [1 favorite]