For Google search by Voice, we used GOOG-411 acoustic
models together with a language model estimated from web query data.
There is a balance to maintain in which the application needs to be compelling
enough to attract users, but not so challenging from a recognition
perspective that it makes too many errors and is no longer useful. Google
makes it easy to push the boundaries of what might be possible while engaging
as many users as possible - partly due to the fact that delivering services
from the cloud enables us to rapidly iterate and release improved versions
of systems
đang được dịch, vui lòng đợi..
