Is speech to text possible?

When will it be possible to calibrate google voice recognition and speech to text?

• When using predictive text you can have it remember your unique words and names. Basically calibration. When I am using Google's Speech to text and have a unique word or spelling of a name, Id like to be able program that over the existing. Another simple example is punctuation. You can say aloud "exclamation point or Question mark and receive "!" and "?" respectively. However Semi-colin, Colin Ampersand and many others only return the word. Another basic function could be calibration for you specific speech patterns. Would work like the Android facial recognition lock screen calibration. Take more photos for a better result. By speaking predetermined words and sentences the system can learn your accent syntax. Im sure linguists could determine the most efficient phrases

• Answer:

  Google is providing only basic speech recognition API and over the years they didn't demonstrate the will to extend or improve their API to support advanced features like vocabulary tuning. Microsoft's speech API is better here, at least it allows you to use grammars. Overall, if you expect additional flexibility from the speech recognizer, you can consider custom solutions like http://cmusphinx.sourceforge.net. The accuracy in recent versions is comparable to the one from Google but in addition you can get very cool features: custom vocabulary with grammar (it will finally recognize the words like Spotify), ability to adapt to your own pronunciation, efficient search for configurable keywords and so on.