By Joe Levi | December 14, 2010 2:16 PM
Google has been doing a lot of work with voice searching lately. We saw a microphone icon in the Google Search widget in very early versions of Android to allow voice searching. We saw that same microphone added to the soft keyboard (IME) to enable text input via spoken words. More recently we’ve seen the rise of “intents” where various naturally spoken phrases are interpreted and turned into actions by Android (like “Send email to Brandon, Subject Awesomeness, Message Hey did you hear the good news question mark Google’s Voice Search for Android can now learn how you talk and get even better at understanding your specific speech traits period Cool period”).
When Google launched Voice Search over two years ago, they “wanted it to ‘just work’ right out of the box, without an initial setup process”. They did this by building speech models that were generic and broad enough to “just work” with lots of people, regardless of age, gender, accents, and other “traits”. Accommodating all these traits, however, sacrifices individual accuracy. Even though, statistically speaking, most spoken content is recognized correctly, the outliers may not feel that it does such a good job recognizing the way they speak.
While Voice Search currently does a remarkably good job, with a recent update, Google hopes people will opt-in to making it even better — at a personal level.
Google decided to build a more accurate voice model per-person by listening to the individual’s voice and learning how that individual speaks. Not only their, they made it an opt-in feature, so Google’s not going to start building a library of how you speak without at least asking for your permission first.
To get started you’ll need to head to the Android Market and download Google Voice Search. Install it, and head to Settings, Voice input & output, Voice recognizer settings, and check the Personalized recognition box. Improvements won’t be immediate, but the more you use Voice Search, the better it will get.
Unfortunately, the personalized recognition functionality is only available for English speakers in the U.S. who are running Android 2.2 or higher (and have the latest version of Google Voice Search installed). They do plan to support other countries and languages in the future. Even if you’re not in the U.S., or aren’t running Android 2.2+, go ahead and grab it. Even without the personalization, Google Voice Search “just works”.

Source: Google Mobile Blog










