I have a very specific production problem. The client has supplied voiceover clips for a video. These were recorded in a home office as 16-bit .wma files using, I believe, a headset, so there are problems. It is not possible to get a redo, but for the sake of my own conscience I want to give the client as good a product as is possible. Audio is not my field, by a long way, but I do have Audition (CS 5.5) available.

The major problems, in order of obnoxiousness, are:

  • Background noise, mainly a low-level 60Hz hum with harmonics and a general mid-range hiss that sounds like an AC running. There is nothing in the recording above about 12k.

  • Low overall recording level (about -20 avg)

  • Variable voice level, requiring compression

  • Muddiness (baritone), requiring EQ in the 100-220 Hz range.

  • Variable vocal tone in places, caused, I think, by head movement relative to the mic.

  • Misc. clicks and pops

I can convert the tracks to 32-bit floating point to gain some flexibility, and after several hours of study online, I think I can probably handle the individual steps of noise removal (multiple steps?), EQ, gain and compression. What I don't know is the best sequence to apply them to have the best possibility of a decent end product, and I've had no success in tracking down any best practices. It seems they must exist somewhere.

  • Could you just re-record them yourself using a decent mic? If they did the VO on a computer headset they obviously don't care much about what the voice sounds like... Maybe they don't care whose voice it is! It would be faster and have better results than trying to polish a t***.
    – Bill Gribble
    Jan 21, 2012 at 19:58
  • Unfortunately not. The VO was recorded a long way from here. It's a matter of the client's budget and inexperience. I'm not trying for a silk purse, but I'd like to get it up to a cotton t-shirt... :)
    – Alan Gilbertson
    Jan 22, 2012 at 4:14

1 Answer 1


I'd start with the hum, as this becomes more difficult after any other editing step. But right at the beginning, you should be able to do it very well with a multiple-notch filter.

Then I'd tackle clicks and pops. For this you get the best results by doing all of them manually, but it's of course very laborous. But fully automatic tools, in my experience, do too much harm to the desired signal. If the events are really loud it should be easy enough to track them in the waveform display, if they are rather subtle if may be better to just leave them there. For pops, a simple low-cut filter should already help quite a bit.

Hiss. Once you've removed the louder other noises, it might be possible to cure this with an ordinary noise gate. If it's but little below the voice level, this won't work as well, but you might succeed with an FFT-based frequency domain noise gate. This might introduce somewhat strange audio artifacts, but it might be worth it.

Only after all the undesired signals are surpressed as far as possible, you should start with bringing out the voice better. Here, you may have to experiment with the order of application yourself, but as a basis I'd suggest compressor->EQ.

  • This. I would just add that depending on the video you could run the vocals tru some fx chain to make the low fi more attractive or less obvious: Short and strong reverb with hi reflections and a analog-feel compressor (like The Glue) would hide some of the bad quality. If it's possible, it would be interesting to add a heavy filter and more compression to make it look like a phone call or something. If you can add background noise, anything with a little bit of white noise will compensate the low quality of the audio - our brain seems to understand better 'broken' audio with white noise.
    – Johnny Bigoode
    Jan 23, 2012 at 15:17

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.