Out of curiosity, I implemented the following headphone amp:
I used the exact same components (yes, even the 2N2222/2907 are the nice metal-can ones), checked all connections several times, but the amp just doesn't work. I'm using a bench power supply at 12V, as recommended, and I'm driving a 180 ohm headphones. The amplifier outputs something, but it is very quiet, the gain is less than 0.01. Practically unusable. Also I tried simulating the circuit in Falstad's online simulator, where it shows basically the same thing. The same circuit is available in a few other sites, so it's probably drawn correctly.
One thing I noticed is that if I turn the supply voltage down, the gain actually increases, to the point that at 6 VDC it's almost unity gain, but with heavy distortion. There's almost no distortion at low levels (e.g., if I'm turning the pot to 1/10), but the distortion is unbearable at any usable gain levels.
After some tweaking in the simulator, I simply removed the biasing diodes D1 and D2 - and voilà - the amp works perfectly. It amplifies around 10x, and the distortion is gone.
Now, what I'm asking here is why are these diodes required, what kind of biasing is this? It is obvious that they do serve a purpose - without them, Q2 goes into saturation, and is basically bypassed - the amplification is done by Q3 alone. Without them, the current through Q2 and Q3 increases several times - at 12 V it is almost 100 mA and the transistors get quite hot (it's bearable at 9V though). The original circuit hints at 15 mA current. Is it possible that the original circuit designer made an error in the specification of the diodes? How can the circuit be "fixed" to adhere to its original idea?
I'm not very adept in analogue circuits, and biasing in particular, so some explanation of why the circuit was designed that way would help a lot.
EDIT: here's a link to the simulation. You can click the switch near the diodes to see what happens.