It looks like the Q1-Q2 thing is a convenient way of biasing up the input transistor's base without injecting power supply noise into the input. That also creates a DC feedback loop: Q1's collector sets Q2's emitter voltage, in turn defining its current by Ve = Ic * (5.6k + 3.4k). Similarly, Q2's emitter sets Q1's base voltage through the R7/R8 divider. (Vb - Vbe) / 10k = Ic, and 12V - Ic * 270k = Vc for Q1. I wonder how sensitive the bias is to scratches and stuff.
The 2SC1815s are 60V parts, so I'd be curious to see if the circuit could be linearized with a higher supply rail. Could be fun to replace R10 with a JFET connected as a current source, and perhaps increase R11 to 1M.
