The idea is to build voice recognition from a large base of samples from different people (male/female, age, accents...).
That's why Google offered GOOG-411, mostly to train their database. And that's why all these solutions include a server-side part, because they need the large sample database that keeps growing.
That's why Google offered GOOG-411, mostly to train their database. And that's why all these solutions include a server-side part, because they need the large sample database that keeps growing.