You would need to isolate the sound you want to use. Adobe Audition is a good choice for this. Take your audio track into Adobe Audition, try to isolate the sound you want (check out tutorials for Audition on how to do this), and then render the result out as a WAV file. Bring that WAV file into AE and get your keyframes from that audio instead. Then mute that layer.
I'll try it!
You generate the changing parameters by either using the Audio To Keyframes keyframe assistant or by using Trapcode Sound Keys (you will have to pay for Sound Keys) to give you a range of numbers. The you typically examine that range of numbers and tie a property like scale to the changing values with an expression. If the property you are trying to modify is an array like scale, then you have to turn the values generated by Audio to Keyframes or Sound Keys into an array. This requires a basic knowledge of expressions.
If you need to separate out specific frequencies or instruments to drive a specific effect you will have to take the audio into an app like Adobe Audition and process the audio to isolate that sound and then render a separate file for that sound. The technique is then the same.
I don't know if an expression written in english will help you but if you are using audio to keyframes and you want to use the Both Channels results to drive scale you can use one of the interpolation functions like linear(t. tmin, tmax, value1, value2) to control scale. The first thing you would do would be to open up the Graph Editor and view the minimum and maximum values for Both Channels. After noting the min and max values you would start an expression for scale. You would then set up the Both Channels as the variable t by typing t = and then using the pickwhip to attach the value of Both Channels and end the line with a semicolon. Next you would set up a variable for scale by typing s = and then open up the Expression Language Menu from the timeline and choose Interpolation>inear(t. tmin, tmax, value1, value2) and then add a semicolon to the end of the line. You would then enter values for tmin and tmax from you notes and then choose values for minimum and maximum scale. You would then set up an array by typing [s, s] to use the interpolated values as scale functions. If you wanted to have the layer bounce back and forth between 50% and 100% while the Both Channels values change from .2 to 1.5 your expression would look like this:
t = thisComp.layer("Audio Amplitude").effect("Both Channels")("Slider");
s = linear(t, .5, 1.5, 50, 100);
I hope this helps you get started.