Speech recognition had long been considered difficult to program, but now appears standard in most modern smartphones.
What’s the secret? Statistical modeling.
While any word can begin a sentence, the words that follow do not appear in a random order. Machine learning can be employed to enable the software to assess the likelihood of a given word based on surrounding words. The software would also have ‘learned’ from vast repositories of digital documents and speech from other users.
Informed guesswork can also be improved by incorporating the user’s personal information and location. However, it remains to be seen how much personal information consumers are willing to give up in return for a more reliable service.
Read the full article here.