Any audio equivalent of "extract a person or object using Photoshop"?
November 19, 2013 8:46 AM Subscribe
I have a sound recording that ideally I would like to remove the background sound from. My recording has unfortunately captured the unwanted sound of an episode of Doctor Who being played from a TV in the background. Is there a way to filter that out?
Based on my googling, this audio separation problem is on the one hand, extremely hard in the general case, and yet on the other, trivial if you have an exact copy of the unwanted sound component. In my case, I am wondering if an actual copy of the TV show audio, even though it is not an identical sound signal, could be used as a "hint" for some special magical software filter or technique. I am not familiar with audio software, beyond having downloaded Audacity once or twice. I viewed some youtube videos demonstrating the Adobe Audition app, but their results seem more geared toward removing short and simple bits of sound, as opposed to a known but complex and long pieces of background sound.
Based on my googling, this audio separation problem is on the one hand, extremely hard in the general case, and yet on the other, trivial if you have an exact copy of the unwanted sound component. In my case, I am wondering if an actual copy of the TV show audio, even though it is not an identical sound signal, could be used as a "hint" for some special magical software filter or technique. I am not familiar with audio software, beyond having downloaded Audacity once or twice. I viewed some youtube videos demonstrating the Adobe Audition app, but their results seem more geared toward removing short and simple bits of sound, as opposed to a known but complex and long pieces of background sound.
Can you describe exactly what's in the foreground (the part[s] of the audio you do want to preserve)?
I agree with frequently -- it may be tough to do anything manually with better results than you'd get from just some simple EQ.
(A clean copy of the show's audio won't be much help in practice - that was good thinking, but in your real-life recording situation, the TV audio will have gone through too many transformations for any meaningful matching-up without creating the artifacts frequently mentions.)
Luckily the bass frequencies of the TV show should be artificially prominent (esp. since most consumer TVs already crank the bass) so just a high-pass filter may help more than you'd guess.
posted by kalapierson at 9:15 AM on November 19, 2013 [1 favorite]
I agree with frequently -- it may be tough to do anything manually with better results than you'd get from just some simple EQ.
(A clean copy of the show's audio won't be much help in practice - that was good thinking, but in your real-life recording situation, the TV audio will have gone through too many transformations for any meaningful matching-up without creating the artifacts frequently mentions.)
Luckily the bass frequencies of the TV show should be artificially prominent (esp. since most consumer TVs already crank the bass) so just a high-pass filter may help more than you'd guess.
posted by kalapierson at 9:15 AM on November 19, 2013 [1 favorite]
If it is a stereo recording you can invert the phase of one side and mix it to mono. If you get the levels right the sounds in the middle of the stereo spread disappear leaving the sounds on the side. Search "how to remove vocals from song" for tips.
posted by bhnyc at 9:24 AM on November 19, 2013
posted by bhnyc at 9:24 AM on November 19, 2013
I have fixed and repaired a lot of sound, with better tools than you have at your disposal. With your level of skill and experience here's what would be possible:
— You could remove a very consistant sound throughout a recording, like a hum or buzz.
— You could minimize a high set of sounds (chirpy birds) or a low sound (truck rumbles) in a recording, but the result will be bassy or tinny, respectively.
@kalapierson is correct. The TV recording won't help that much. It's not a clean copy, because it doesn't take into account how the TV sound is being affected by your room, distance to the mic, etc.
The best you can do is, if the TV sound is all low and rumbly and in the distant background, minimize the lower frequencies in an EQ setting.
If you want to try with the easiest consumer tool for this, it is in the audio program that comes with Final Cut Pro — Soundtrack Pro. They have a noise reduction tool that is very simple (as these things go). If you can borrow a copy of the program that will still not give you perfect results, but let you see at least what is possible.
posted by amoeba at 10:07 AM on November 19, 2013
— You could remove a very consistant sound throughout a recording, like a hum or buzz.
— You could minimize a high set of sounds (chirpy birds) or a low sound (truck rumbles) in a recording, but the result will be bassy or tinny, respectively.
@kalapierson is correct. The TV recording won't help that much. It's not a clean copy, because it doesn't take into account how the TV sound is being affected by your room, distance to the mic, etc.
The best you can do is, if the TV sound is all low and rumbly and in the distant background, minimize the lower frequencies in an EQ setting.
If you want to try with the easiest consumer tool for this, it is in the audio program that comes with Final Cut Pro — Soundtrack Pro. They have a noise reduction tool that is very simple (as these things go). If you can borrow a copy of the program that will still not give you perfect results, but let you see at least what is possible.
posted by amoeba at 10:07 AM on November 19, 2013
If the TV noise is fairly quiet, you might be able to use a noise reduction filter to make things at least a little better. Audacity can do that (I have no idea how well it compares to any other software in this respect). I don't remember which menu it's under - possibly Effects? It might be a problem if you don't have a small bit of audio to select where the only thing you hear is the TV noise; the noise reduction dialogue asks for that in order to build a noise profile.
posted by trig at 10:21 AM on November 19, 2013
posted by trig at 10:21 AM on November 19, 2013
If the audio track is not sensitive (or you trust an internet stranger) post it for me and I'll play a little. I use various professional tools (pro tools, sound forge etc) and have the chops to use them. Frequently and Kala are both spot on, but I'm happy to give it a go for you if you'd like. Send the link via memail and I'll see what I can do.
posted by chasles at 10:37 AM on November 19, 2013
posted by chasles at 10:37 AM on November 19, 2013
It might also help to know what you do want to preserve, Is it voice, music etc?
posted by TwoWordReview at 11:50 AM on November 19, 2013
posted by TwoWordReview at 11:50 AM on November 19, 2013
I will also volunteer time to do this. I did a similar thing for a Redditor once here.
Unlike some in this thread, I feel that if you have a clean copy of the sound you want to remove, and perform EQ on that sound first before trying the inversion, you'll have decent results.
It won't be perfect, but if it's not for professional use it should be more than passable.
posted by tomierna at 12:20 PM on November 19, 2013 [1 favorite]
Unlike some in this thread, I feel that if you have a clean copy of the sound you want to remove, and perform EQ on that sound first before trying the inversion, you'll have decent results.
It won't be perfect, but if it's not for professional use it should be more than passable.
posted by tomierna at 12:20 PM on November 19, 2013 [1 favorite]
Best answer: I think this situation is a little different though because the audio from the TV will have been filtered by the room where the audio was recorded.
If I was to try that method I'd try and capture the impulse response of the room/mic combo and convolve that with the clean audio first. (Use the same mic in the same position as the mixed audio recording and stand in front of the TV and clap your hands to record the impulse)
Since you won't get sample to sample match-up for the phase inversion technique I'd then use the knockout plugin to do something similar in the frequency domain.
So yeah, I think it's a little more complicated but you could get decent enough results.
posted by TwoWordReview at 12:36 PM on November 19, 2013
If I was to try that method I'd try and capture the impulse response of the room/mic combo and convolve that with the clean audio first. (Use the same mic in the same position as the mixed audio recording and stand in front of the TV and clap your hands to record the impulse)
Since you won't get sample to sample match-up for the phase inversion technique I'd then use the knockout plugin to do something similar in the frequency domain.
So yeah, I think it's a little more complicated but you could get decent enough results.
posted by TwoWordReview at 12:36 PM on November 19, 2013
I'd also say that you probably wouldn't be able to subtract the soundtrack from the recording for a whole bunch of reasons. Certainly whenever I've tried to do anything like that before I've wasted a lot of time for nothing.
Using Izotope RX's spectral retouching tool might give the best results, though still not perfect. I've used it to remove all sorts of sounds in the past. It depends how much of the recording there is and how pristine a result you want.
posted by Grangousier at 12:58 PM on November 19, 2013
Using Izotope RX's spectral retouching tool might give the best results, though still not perfect. I've used it to remove all sorts of sounds in the past. It depends how much of the recording there is and how pristine a result you want.
posted by Grangousier at 12:58 PM on November 19, 2013
In Adobe Audition you can "paint out" unwanted sounds, It's not perfect, nor is it easy. It works best on unwanted sounds that occur in the clear between sounds you want to keep. If the sounds are simultaneous or overlapping, you can have some success at removing the unwanted ones if they are in a fairly discrete frequency band.
Like I said, it's difficult to get the hang of the techniques, they don't always produce expected results and you can introduce unwanted effects into the remaining sound. It would be a huge PITA to do this on an audio recording of any length with constant unwanted sound. It's more for removing an occasional stray cough or tongue click.
And yeah, the thing with having an exact copy to invert and cancel the unwanted sound is pretty dicey, and requires a copy that's pretty close to sample-accurate in order to get good results (based on my limited experience with that technique). If you do go for that technique, you want to invert the waveform, not phase-shift it.
posted by under_petticoat_rule at 8:24 PM on November 19, 2013
Like I said, it's difficult to get the hang of the techniques, they don't always produce expected results and you can introduce unwanted effects into the remaining sound. It would be a huge PITA to do this on an audio recording of any length with constant unwanted sound. It's more for removing an occasional stray cough or tongue click.
And yeah, the thing with having an exact copy to invert and cancel the unwanted sound is pretty dicey, and requires a copy that's pretty close to sample-accurate in order to get good results (based on my limited experience with that technique). If you do go for that technique, you want to invert the waveform, not phase-shift it.
posted by under_petticoat_rule at 8:24 PM on November 19, 2013
Response by poster: I got the "Kn0ck0ut" VST plugin to work and it did the job admirably. Thanks, everyone!
posted by polymodus at 12:49 AM on November 22, 2013 [1 favorite]
posted by polymodus at 12:49 AM on November 22, 2013 [1 favorite]
This thread is closed to new comments.
for instance if the sound you want is one with a small frequency range, you can create an EQ filter so that all the undesired frequencies are dramatically lowered, and the desired ones are boosted,
it's quite difficult to do perfectly, and in most cases there will be some artefacts left over,
perhaps your best best would be to use a program like spectra layers pro, where you see a visual representation of your sound, and you can erase all of the frequencies around the sound that you don't want,
theres kind of a steep learning curve with this type of thing, but there will be tutorials online, adobe audition has a spectral editor i think, theres a program called spectral layers and another called izotope RX, they will all be available with trial periods if its only this one file that you want to fix,
my approach in these situations though is to use eq just to minimise the noise and accentuate the desired sound as much as possible because i find the sound of spectral editors, or other noise removal software most of the time to be more unnatural and disruptive than to keep a coherent sound and try to minimise or mask the offending noise
posted by frequently at 9:06 AM on November 19, 2013 [1 favorite]