Modems under Lab Conditions
The following sounds were recoded between modems in a 'lab' condition - that is, operating on an isolated current loop and not on the real telephone system. All recordings are in WAV to ensure the highest quality. Note that since all modems are analog modems, there cannot be any simulated 56k lab condition connections given my equipment.
This first set of sounds were recorded between two Conexant or Rockwell Hardware Controller Modems. Notice both modems are negotiating V.8bis for almost all modulation speeds which is not really appropriate as older modems did not have this ability. Thanks to the use of the AT+MS command, we can force almost all types of modulation supported and get what we want. An exception to this is the Bell 103 and Bell 212A standards where the modems failed to make a connection despite being told to do so.
Rockwell
And here is a conversation between a US Robotics Message Modem and the Rockwell modem above. Notice the V.8bis negotiation isn't happening because these modems do not negotiate V.8bis. Also notice some different approaches to connecting, and also the fact that Bell 103 and Bell 212A function correctly. There is also an example of null speed shift and retrain - note that this requires the speaker to be on while connected (i.e. atm2) in order for you to hear this happening.
US Robotics
I was going to record an Ambient talking to a Rockwell, but the DAA circuit in the Ambient modem doesn't like my synthetic loop current generator and decides to short it out resulting in no connection - but that gives us a chance to observe some "connection recovery" principles (as specified by the ITU) going on. Actually, on further examination, it seems the hybrid in the modem was saturated by the loop current somehow - resulting in signal distortion.