How to lower peaks/raise valleys on an audio clip?
June 20, 2011 1:22 PM Subscribe
I'm a journalist; I record all my phone interviews, outputting them as WAV files. I just did one, and for reasons I can't determine, the audio levels are haywire - my input (when I'm asking questions) is eardrum-breaking loud; the answers to the questions are whisper-quiet.
I'm trying to figure out how to remedy this automatically; flatten the peaks and boost volume on the whole thing; or flatten the peaks and raise the valleys (or another solution?) I'm on a Mac.
A compressor! You've got one in Garageband. It makes the louds softer and the softs louder.
posted by Jon_Evil at 1:32 PM on June 20, 2011
posted by Jon_Evil at 1:32 PM on June 20, 2011
A compressor doesn't make the soft sounds louder unless you also normalize.
Compression just 'flattens' a sound by lowering the volume of anything that's over a certain level.
Normalization raises the volume until the average volume of the sound is at a certain level...
The combination of the two raises the volume to a minimum level, while smooshing down anything that's too loud, so you don't get clipping.
You do get distortion from the process, though.
In the case of an audio interview where the two sounds are separate in time, I think the best option would be to adjust the volume of the sections by hand.
posted by empath at 1:36 PM on June 20, 2011
Compression just 'flattens' a sound by lowering the volume of anything that's over a certain level.
Normalization raises the volume until the average volume of the sound is at a certain level...
The combination of the two raises the volume to a minimum level, while smooshing down anything that's too loud, so you don't get clipping.
You do get distortion from the process, though.
In the case of an audio interview where the two sounds are separate in time, I think the best option would be to adjust the volume of the sections by hand.
posted by empath at 1:36 PM on June 20, 2011
I think The Levelator will do this for you, although it might also raise background noise to an annoyingly high level.
posted by griseus at 1:43 PM on June 20, 2011 [2 favorites]
posted by griseus at 1:43 PM on June 20, 2011 [2 favorites]
I don't know if Audacity has it, but Cool Edit Pro (and I assume Adobe Audition) has a 'clip restoration' function which attempts to restore the clipped peaks. Results will vary depending on the amount of clipping present & the type of audio, but it's usually pretty good on semi-severely clipped voice. It'll certainly take away much of the harsh harmonics caused by the squaring off of the waveform. Drop the levels by ~9dB, then run clip restoration with a target of -2~3dB.
And yeah, then adjusting by hand - you could faff around all day re-adjusting levels and tweaking a compressor to work, but running a light de-noiser, selecting sections, and normalising those individually will almost certainly give a much better result.
posted by Pinback at 4:31 PM on June 20, 2011
And yeah, then adjusting by hand - you could faff around all day re-adjusting levels and tweaking a compressor to work, but running a light de-noiser, selecting sections, and normalising those individually will almost certainly give a much better result.
posted by Pinback at 4:31 PM on June 20, 2011
I am a professional sound engineer.
Nth-ing doing it by hand on a clip by clip basis. Yeah, it sucks, but anything involving compression and normalization will make your voice sound like poopey-doops.
posted by aloiv2 at 9:23 PM on June 20, 2011
Nth-ing doing it by hand on a clip by clip basis. Yeah, it sucks, but anything involving compression and normalization will make your voice sound like poopey-doops.
posted by aloiv2 at 9:23 PM on June 20, 2011
My first suggestion would be to find a different way to record the audio. If it is just a digital recorder next to a speaker phone, stick the recorder right next to the speaker. Tape it onto the speaker if you have to. (I would note that this is going to sound like ass no matter what you do, especially if the room you are in is echoey.) There should be adapters available that you can plug in between the handset and the base of a desk phone, or between a cell and a headset. This would be optimal.
My second solution would be to:
1- Look at the waveform. Are the peaks of your voice clipped? If that's the case, you need to reduce the volume you are recording at, because you really can't clean those up once they are clipped. Even if you reduce the volume, the clips will just be quieter, and still "sound" loud. Digital recording is all about NEVER clipping. In the analog world, you can overdrive the input a little and it usually will sound OK. But in digital, it simply cuts off and you lose information.
(The way around this is to record at a fairly deep bit depth, and shoot for say -6db as your zero point. Any surprises still have some headroom to work with. Then you clean it up in post.)
2- Once you've got a clean recording, set up a (dynamic?) compression filter. I'm not sure about audacity, but in Cool Edit (or Audition), you can program the compressor with rules. If you look at the waveform, you might see that the other person's voice never goes above -16db. So you set your compressor to (for example) reduce anything that's above -12db by 12db. That *should* get the waveform looking pretty flat, and then you normalize it to -1db.
3- Or, look for a compressor plugin that is something like "FM Radio" or "broadcast post-processor" or something like that. That should have the settings built in to make the final product good enough. Or maybe do step #2, and then run it through this one.
(Technical note: most compressors I've seen work in both directions. They will both boost and reduce. Also, normalizing (on Audition) doesn't work on the average, it works on the peak(s). It finds the peak, and amplifies the entire selection by the difference between the value of the peak and the normalize to value.)
posted by gjc at 6:00 AM on June 21, 2011
My second solution would be to:
1- Look at the waveform. Are the peaks of your voice clipped? If that's the case, you need to reduce the volume you are recording at, because you really can't clean those up once they are clipped. Even if you reduce the volume, the clips will just be quieter, and still "sound" loud. Digital recording is all about NEVER clipping. In the analog world, you can overdrive the input a little and it usually will sound OK. But in digital, it simply cuts off and you lose information.
(The way around this is to record at a fairly deep bit depth, and shoot for say -6db as your zero point. Any surprises still have some headroom to work with. Then you clean it up in post.)
2- Once you've got a clean recording, set up a (dynamic?) compression filter. I'm not sure about audacity, but in Cool Edit (or Audition), you can program the compressor with rules. If you look at the waveform, you might see that the other person's voice never goes above -16db. So you set your compressor to (for example) reduce anything that's above -12db by 12db. That *should* get the waveform looking pretty flat, and then you normalize it to -1db.
3- Or, look for a compressor plugin that is something like "FM Radio" or "broadcast post-processor" or something like that. That should have the settings built in to make the final product good enough. Or maybe do step #2, and then run it through this one.
(Technical note: most compressors I've seen work in both directions. They will both boost and reduce. Also, normalizing (on Audition) doesn't work on the average, it works on the peak(s). It finds the peak, and amplifies the entire selection by the difference between the value of the peak and the normalize to value.)
posted by gjc at 6:00 AM on June 21, 2011
will make your voice sound like poopey-doops.
Is that the technical term?
posted by empath at 6:02 AM on June 21, 2011
Is that the technical term?
posted by empath at 6:02 AM on June 21, 2011
This thread is closed to new comments.
Then you have two options -- you can adjust the volume by hand for each section (probably won't be very hard to see the difference).
Or you can compress and normalize, but you're going to get some distortion doing it that way.
posted by empath at 1:32 PM on June 20, 2011 [1 favorite]