there doesn't seem to be any obvious reason. Rg and Rv should be equal, which biases the input at 1/2 Vcc. it's possible that low values load down the audio source, although a 5k input impedance should be high enough that just about anything should work as a source without getting loaded down. you should also verify that the DC voltage on pin 6 of the comparator is also at 1/2 Vcc, and that the triangle wave is symmetric with respect to 1/2 Vcc. it's possible (and i've run into something similar before) that with 10k resistors there, you are leaking current out through the electrolytic (which for this circuit where you have 1/2 Vcc on one side of the cap, the cap is backwards, and will leak current). with the 100k resistors, the - side of the electrolytic is at +2.5V, but the amount of current that can get through is limited by the 100k resistors.