Nope it will really suck. It will not do well with multiple voices, especially because the voices will be different sounding. Moreover, people on telephones do not record well in any case, because telephone equipment provides very low quality sound reproduction, and people talk too fast or to slow and without proper diction, clarity, enunciation, etc., for speech recognition to work well. People run sentences together, change their pacing and tone, change their volume and pitch,and the like. the recording will make a poorer input for speech recognition software to decipher and transcribe. I predict serious disappointment with this procedure.

I presume recording conversations like that is either legal in your state, provided at least one party to the conversation consents; or that you obtain the consent of all parties involved. In my state (Michigan) recording a person without his permission is a felony, even if you are a party to the conversation.

Just my take. Best Regards. ////// Richards //////
============================================

In addition to what Richard points out, the main problem with Multiple voices is that SR is designed for dictation, or structured speech, not for conversational speech. Aside from the fact that at least one speaker will use speech patterns unknown to the SR application (i.e, untrained), the fact that these recordings will be "conversations" will produce very poor recognition results regardless of any other factors.