What latency is required to make software fax using only speaker and mic?
Fabrice Bellard (yes, the QEMU and Javascript-PC-emulator guy, he's awesome ;) ) did some work on software modems. According to his project page, there is code for "soundcard interface" as well as modulation/demodulation of various protocols. I haven't looked at the source, but this might provide useful pieces. There is also the linmodems.org site with further pointers.
Linmodems.org links to some software by the late Tony Fisher, who seems to have had a complete implementation of the sending part of what you want to achieve, although for an ancient workstation. Have a look at the manpage.
Additionally, there is this post on the GNU radio mailinglists from 2003 of a guy that seems to want the same thing, although he didn't have xkcd to describe it back then ;)