Back in your 1st post you said it was working, so you can ignore all of this this if you wish...
Something like the VU meter but with One LED Only... I am hooking just a LED to PWN to sync to speaker value...
A "trick" I use with my lighting effects is to grab a reading once per second and put it into a circular buffer (like the smoothing example). It's a 20 element array, so it contains information for the last 20 seconds. Depending on the lighting effect, I either use the average of the array or the peak in the array as a reference. With this as a reference, the effect automatically calibrates for volume control changes or loud & quiet songs.
My simplest effect sets-up a (usually random) on-off pattern in an array of LEDs. When the loudness is greater than average, the LED states toggle. When it's less than average, the LEDs go back to their normal state. This gives a nice flickering effect with lots of "LED action" no matter how loud I'm listening or how loud the song is. This also works with one LED, where the LED flickers-on above average and turns-off below average. Or, the logic can be reversed to turn the LED off with loudness. Or, it works great with one incandescent flood light, or a pair of floods that toggle oppositely.
I have a VU meter effect using the same LED array (and the same Arduino... it's just a different function in the same sketch). This time the maximum value in my 20-second array is the reference for the "top" of the meter. Again, lots of "LED action" no matter the volume setting or the loudness of the song. But, it makes a terrible "meter" because the sensitivity/calibration is constantly changing. 
To get the widest-possible dynamic range, I also switch automatically between the 1.1V reference and the 5V reference.
I use a [u]peak detector[/u] on the input. It works like the diode/capacitor/resistor circuit that russellz described, except the op-amp compensates for the ~0.7V drop across the diode, and this allows me to use a line-level signal, even a rather low-level volume-controlled line-level signal.
A peak detector (with or without the op-amp) converts the AC audio into a more-slowly changing DC voltage.
The peak detector throws-away the negative half of the waveform. I also have a "protection diode" to make sure I never feed more than 5V into the Arduino.)
By "conditioning" the signal with the peak detector, I can sample the DC waveform about 10 times per second instead of sampling the audio waveform thousands of times per second. It also means that the data in my 20-second array represent momentary peaks and I don't have to worry about the "dips" in the normal audio waveform, or the missing negative-half of the waveform.
Since I'm not using PWM (or dimming), I've got a minimum on-time which is 1/10th of second for the incandescent lights. I think I used 50mS (1/20th of a second) for the LED effects. That's so if the light/LED flashes-on momentarily, it's always on long enough to be visible.