You have to pick a sound sensor that will report back amplitude, like a simple electret mic. Sound levels are usually measured in dB (decibels), which are a little complicated to understand if you are weak in math. Essentially, a reference level must be chosen, which is called 0 dB. That stands for a certain max. pressure level in the sound waves impacting your sensor. Now for every increase of that pressure by a factor of 10, the sound level measurement increases 10 dB. So 20 dB is 100 times the pressure level of your reference value.
As you can see, it aint simple. If I were trying to control speed, I'd just experiment, taking mic voltage level and relating it to a speed. Choose a set of discrete increasing voltage levels, and choose the speed you want to try for each level. For speeds in between these level, pick the upper or lower bounding level to determine the speed. Adjust until you are happy with the performance.
Of course, a raw electret mic only produces a few millivolts, and that's not enough to get good reading from an analog input pin. You can either use an opamp to bring it up to, say. a 0 - 1 v. range, or, much easier, buy an electret mic with amplifier (about $5) from either ADAfruit or Sparkfun (don't remember which).
I think this is the kind of thing you want. Sound Sensing board It has an analog signal that is a measure of the amplitude of the sound. You could read this in with analogRead.