Machine learning has helped Google reduce the percentage of incorrect voice recognition to 4.9%
According to statistics, people are increasingly using the possibility to specify a search query voice. And not only in the case of mobile devices, where often it’s just more convenient, but when using a full PC.
At the event Google I/O, the search giant boasted that the percentage of erroneous voice recognition for the year decreased by almost half. If in July last year, the rate was 8.5%, now it dropped only to 4.9%. Indeed, those who often use the capacity of Google search engine may have noticed that even with the best diction, the noises around and the presence of a sufficiently complex words often the system is completely correctly recognizes spoken.
This not only improves the perception of the use of the service, but also contributes to the development of the voice assistant, as it is completely dependent on the correct recognition of speech.
Improvements associated with the machine-learning techniques. Probably the same modules TPU made a contribution.