[akvadrako]

What you might be able to do is your the text codec as the first pass, then augment the audio with Codec2 or so to capture the extra information (inflections, accent, etc...), for something in between 2 and 700bps.