Audio enhancement question

March 9, 2005 8:38 AM

What are the best tricks, using an audio program like Audacity, for improving a recording of a phone interview where the other guy sounds really quiet, but your voice is loud? What compression/decibal settings are typically best, and what other effects will bring the quiet voice up -- perhaps without making the loud voice much louder?

Not looking for broadcast quality sound here, just trying to make a faint little voice a little easier to transcribe.

posted by inksyndicate to computers & internet (9 comments total)

Not to snark, but have you tried different compression settings? You can just select a small part of the file to experiment on, and undo what doesn't work.
posted by squidlarkin at 8:47 AM on March 9, 2005

P.S. :)

I'm basically wondering if there is anything other than compression that anyone recommends.
posted by inksyndicate at 8:50 AM on March 9, 2005

no just use compression. there's no way to recommend values because they will depend on the difference in volume between your samples.

perhaps try this: try to eyeball the average dB level of the loud voice. Do the same for the quiet voice. Use the compressor to take down anything over (volume of low signal+half the difference) by more than 6:1.

then boost the entire signal by the difference.

blindly, that sounds like it may work. ymmv!
posted by plexiwatt at 9:01 AM on March 9, 2005

Well, you can normalize the quieter parts of the conversation... compression can be really ugly and tricky on speech.

On preview: yes, plexiwatt's got the right idea. The really tricky bit is setting the release time of the compressor.
posted by dirtynumbangelboy at 9:02 AM on March 9, 2005

I am not familiar with Audacity, but if you can select the part of the wave form that you want to make louder, you can just increase the gain (might be called "volume") on those portions you want to make louder. You could also select the quiet bits and use "normalize" to maximize the volume without clipping. Neither of these methods require compression, but results may vary. These methods may also take more time since you have to select each bit of phone-talker's voice and treat it separately.
posted by turtlegirl at 9:07 AM on March 9, 2005

Also, I second plexiwatt's suggestion for only compressing the loudest parts of your own voice.
posted by turtlegirl at 9:09 AM on March 9, 2005

It's been a while since I used Audacity, but depending on the length of the file, I think you may get the best results if you manually select the quiet portions and amplify them.

IOW, what turtlegirl said.
posted by recursive at 12:50 PM on March 9, 2005

In Audacity, you highlight the too-quiet part, go to the Effects menu, and select Amplify.
posted by Miko at 1:03 PM on March 9, 2005

I like to do compression manually if i have a nice visual waveform display. for non music audio its a piece of cake First I remove little inaudible spikes or huge short dynamic changes (like pops) that eat up headroom and remove them. Then I'll go through and use dynamics processing to trim the peaks of the loudest stuff and normalize (raise level of highest peak to 0.0) some of the quiet stuff. then I'll normalize the whole wave. You start to get a feel for how much you can do with the dynamics without audible changes and when it becomes obvious. I can generally do a decent scratchpad job of smoothing levels, raising headroom, and improving legibility of a recording without ever listening to it just by looking at the waveform.
posted by 31d1 at 11:24 AM on March 10, 2005