Speech recog. software is so unreliable as everyone speaks differently. If you could look at speech patterns visualy like handwriting, you would see everyone's voice is different, but recognisable as intelligent speech.
The only problem is with speech very has different dialects, speech styles, lispts, etc. Handwriting is pretty much identical to each persons (compared to speech) that's why it's so hard to make a reliable speech recog. program. It has to be set up to each person individually. People who are disabled and need a speech recog. use very expensive programs, as they the government pays for it.