Artificial intelligence is a field as wide as the endeavours of humankind itself. One of the best ways to handle such a complex subject is to find niche areas that can be investigated in depth. Then, at a later point, to abstract away from these areas to find patterns of common solution.
Speech recognition is one such area. Microsoft Research has recently announced progress in building a general-purpose speech recognition capability – which does not require the hours of training typically needed to adapt the software to your way of speaking.
Their approach, presented during Interspeech 2011, breaks down the problem to use the finer-grained yet more numerous senones, rather than the traditional phonemes. This has led to an improvement in word-recognition of more than 30%.
Read the article at Microsoft Research here: Speech Recognition Leaps Forward