If memory serves me correctly, the audio bandwidth in both cases is just 8kHz - enough for voice transmission over the telephone network.
The audio can be encoded/decoded using either CVSD or PCM using logarithmic a-law or μ-law quantization. See links below for an explanation.
I believe it to be unlikely that you will ever determine which of the above schemes is in use for a module type Bluetooth solutions because I think the decision which quantisation and encoding/decoding is decided as the wireless link is created and no shared with the outside host microcontroller. Thus you need to just determine which modulation scheme works best after trying a handful of different solutions. If you run an actual Bluetooth stack on your PC/microcontroller, you are more likely to be able to decide which one to use, but you are still limited by the device you choose to connect to - if it is an off-the-shelf consumer headset, the headset may only support one type and that it the one you are stuck with!
If you search around on the internet for "Bluetooth Module HSP" you will find plenty of vendors offering complete modules and kits that can be used to create a Bluetooth based Headset or Handsfree link. The usual suspect (Digikey, Element14/Farnell, RS etc.) all stock various kits as do many on-line dealers.
In terms of considering a modulation scheme, I would start by considering the DTMF tones that you can generate with any modern telephone (see here http://en.wikipedia.org/wiki/Dual-tone_multi-frequency_signaling). I doubt your overall data throughput will be very high, but I have seen many projects that used DTMF signalling over telephones to control electronic systems - simple stuff like turning lights on and off etc.
Hope this provide some helpful hints.