I am working on compressing speech, and I am finding that the minimum attack time is too long – I can’t get the slide to go below 0.1 second. Derry’s book PC Audio Editing for example suggests 50ms (yes milliseconds) as a good starting point. I am finding that the entrances to a new line of conversation are not compressed at all because the attack is too slow, leaving high entrance peaks that are hard to get rid of. I’m using hard limiting to get around it, but does anybody have any other ideas?
What you really need is something called “lookahead” where the compressor “looks ahead” by a certain number of milliseconds and preemptively applies the compression. Unfortunately the compressor in Audacity does not have this feature. There is a compressor available (at least in the beta version) called SC4 - this is a more advanced compressor and allows the attack to be as low as 1.5 milliseconds. You will still have the same problem, but for a much shorter time, and correcting this with the hard limiter will be highly effective.
The only real solution is to use a compressor that has “lookahead” - you could try using a VST compressor (using the VST-bridge).
Thanks Steve – it is working great! SC4 has really nice features. Do you have any experience with the setting values for the “knee radius”? I can see that it is is dB, but I’m not sure what range of values from 1 to 10 is considered hard, soft, middle. Any tips?
Thanks again!
Glenn Harvey 
The difference is pretty subtle, but the default of 3.25 works pretty well.
For just giving music a bit of a gentle boost, try something like this as a starting point:
RMS/peak: 0
Attack time: 5
Release time: 50
Threshold: -6 (assuming that the audio is already normalised - if not, set it to about 6 dB below the highest peak)
Ratio: 6 (depends on how much compression you want - higher values for more compression)
Knee radius: 3.25
Makeup Gain: 0 (much easier to normalise after compressing than trying to guess how much gain to apply)
Thanks again! Playing around for voice compression, I can see that RMS 0 really works, but I don’t understand why. Can you help me understand what RMS means or does? I have googled RMS in compression, and I can’t find any technical descriptions.
Thanks again for taking the time to respond.
–Glenn Harvey
RMS - “Root Mean Square”
For a technical explanation see http://mathworld.wolfram.com/Root-Mean-Square.html
but in layman’s terms, it’s a way of taking an average values from a waveform. If you have a sine wave that goes up and down from +1 to -1 , the RMS value is about 1.4
The sense of “loudness” that we detect with hearing corresponds much more closely to RMS measurements than to peak measurements, so for things like compressing vocals, using RMS values produces a more natural result.
Peak values are useful if you are applying heavy compression to limit sudden “spikes” (peaks) in the sound - this is sometimes called “hard limiting”.