What speech recognition algorithms are used by Google?

It is all pretty standard - PLP features, Viterbi search, Deep Neural Networks, discriminative training, WFST framework.If you want to study modern speech recognition algorithms, I recommend you to read the following well-written book: http://www.springer.com/us/book/9781447157786 If you are interested in Google's inventions, check http://research.google.com/pubs/SpeechProcessing.html, they describe everything rather accurately.Most of the algorithms Google uses are available in open source, for example in http://github.com/kaldi-asr toolkit. There are big datasets available. So open source can actually compete pretty well with Google, in particular in specialized domains. If you want generic recognition, it is hard to beat them, but for many cases like address query or lecture transcription on specific topic like biology you can easily get way better accuracy than Google provides.Google does have a very big dataset for training, they experiment with hundreds of http://research.google.com/pubs/pub41133.html of speech and terabytes of collected website data. They put effort in large data collection and manual verification. Data is what makes their recognizer good, but algorithms are also important. Google has brilliant speech team and that team ensures leadership in the area. There are competitors from Apple or Amazon though, as well as open source competition.