Puzzler: Transformer hum with soft-start modification on GFA-565

Hey all, I can't seem to figure out this transformer humming problem, and could use some advice...

I'm modding an Adcom GFA-565; installing two 1000VA transformers in place of the original 1250VA. (One transformer for each rail.)

Trouble is, the new transformer does not have a 100V primary tap, and the soft-start circuit on the Adcom is designed to work off of this tap, using the transformer primary as an auto-transformer. The circuit is simple. A 24V relay is driven by a rectifier, resistor and filter capacitor. The resistor is 680 ohms, which drives the 475 ohm relay at about 24V.

The relay pulls in after about 100ms to short out the 4.7 ohm resistor in series with the transformer primary.

So, without this 100VAC source available, I thought I could just use hot 120VAC and increase the power supply series resistor value. I calculated 880 ohms for R502 and 5K for R501.

The circuit behaves exactly as intended with the new 120V levels, just with increased dissipation in the resistors. (I've doubled their wattage ratings.)

The soft-start works beautifully. The voltage ramps up at almost the same rate as the original transformer, and the relay clicks in to short the 5ohm padding resistor at about 60V. Good to go.

But the transformers HUM! Ugh. Why?
I feel like I'm missing something basic here.

Funny thing is, they don't hum right away. It takes about 10 seconds after power-on, and then the hum starts to fade in.

If I power the soft-start circuit with a separate 120V isolation transformer, there is no hum. Beautiful.

Why does connecting it to mains versus the 100V tap matter?

Schematic attached, and red lines drawn how I have modified it.

Thanks for looking!

modified-gfa-565-soft-start.png
 
So, I'm not sure I'll ever know what caused this, but I have a workaround. Apparently some isolation is needed. I've installed a small 28V transformer to drive just the relay portion of the circuit, breaking the trace just before D401, and leaving the 120VAC to drive the opto-isolator bias switch. R502 is changed to just 40ohm, so it does not heat up like the original 680. C501 is now 1000uF, which gives a time-constant for the relay to click in that's close to the original.