Here's a table I found from Tyco Electronics that details typical contact materials, the melt voltage, the arc voltage, and arc currents:
As you can see, we may be in the optimum range with all of these at 24V - some of them even at 12V (I don't know what the current is, either, but it doesn't look like it takes very much.).
Tyco describes the use of the protection diode (1N4148 in previous designs) and its result in holding up the relay coil:
"Some relay users connect a diode across the inductive load to prevent
counter voltage from reaching the contacts. When the relay contacts
open, the stored energy of the inductance recirculates through the diode,
not through the arc. While this is an acceptable method of protecting the
contacts, it does result in lengthened hold-up time of the inductive load."
So maybe the use of the suppressor diode is what speeds things up?
Thanks for the info, Tom! You might find this interesting.
I carried out some tests to see just how quickly the relays could be operated. The results were something of an eye-opener (and I knew about the added delay caused by a diode!). The relay I used was a small 24V coil unit, having a 730 Ohm coil and with substantial contacts (at least 10 Amps). With no back-emf protection, the relay opened the contacts in 1.2ms - this is much faster than I expected, but the back-emf went straight off the scale on my oscilloscope, and I would guess that the voltage was in excess of 500V. When a diode was added, the drop-out time dragged out to 7.2ms, which is a considerable increase, and of course there was no back-emf (Ok, there was 0.65V, but we can ignore that). Using the diode / resistor method described above, release time was 3.5ms, and the maximum back-emf was -30V, so this seems to be a suitable compromise.
I was not able to test the zener method prior to publication, since I did not have the 24V zeners needed on hand. I would expect this scheme to be as good or better than the diode / resistor combination. The graphs below show the behaviour of the circuit with and without the resistor and diode. The estimated 500V or more is quite typical of all relays, which is why the diode is always included. This sort of voltage will destroy most transistors instantly. It is exactly the same process used in the standard "Kettering" ignition system used in cars, but without the secondary winding, or the "flyback" transformer used in the horizontal output section of a TV set.
Figure 4 - Relay Voltages
The trace labelled 'Contacts' is representative only, and is not to scale. The peak relay voltage (above left) exceeded my oscilloscope's input range (and I was too lazy to set up an external attenuator), and as shown is cut off at my measurement limit. I estimate that the voltage is greater than 500V.
Note that the kink in the relay voltage curve is caused by the armature (the bit that moves) coming away from the relay pole piece, and reducing the inductance. This causes the stored magnetic charge to try to increase the voltage again, but it is absorbed by the resistance and dissipated quickly. The contacts open at the point where the previously closed magnetic field is opened as the armature moves away from the pole piece. As can be seen, this is 3.5ms after the relay supply is disconnected.
These graphs are representative only, as different relays will have different characteristics. As noted above, I cannot predict what sort of relay you will be able to obtain, but the behaviour can be expected to be similar to that shown. All tests were conducted using a 24V relay, having 10A contacts. Upon contact closure, I also measured 2.5ms of contact bounce. Provided your amplifier is stable by the time the contacts close, this will be completely inaudible.