I find it difficult to believe that we are still unable to produce software that can reliably translate speech into text for the deaf.
But initial research shows they can only be trained for one voice and will produce gobbledygook in a meeting setting though a journalist for the Guardian Newspapers managed to produce something close to legible:
However, a deaf user isn't trying to train his/her voice to use these products, we are trying to train the voice of others and this is where these products fail.
Technology Research News (Via Slashdot) reports that some researchers from Carnegie Mellon University are working on a PDA which can translate speech. They have produced a proof of concept which shows that such technology is possible but the PDA, called a Speechalator is not due to be finished until 2008 and it was designed mainly for the 2008 Olympic Games and at the moment, it translates speech from one language to another though I hope it would benefit the Deaf community too. It shouldn't be a problem to configure it to output text instead of speech.
Whether the Speechalator is a success or not, it has one obvious advantage over iCommunicator, Dragon Naturally Speaking and IBM ViaVoice in that the speech recognition engine has been optimized to handle spontaneous speech, therefore, it should recognize other voices apart from the owner without any training.