It can really feel like you’ve stepped into the future when you pick up your smartphone, hold it to your face and search for something simply by speaking. When you ask for “car parking” with a thick Boston accent, however, Google might think you’re searching for the late great Tupac Shakur and the future suddenly seems a little less bright. Luckily, Google has introduced personalized voice search, which hopes to account for any number of variations in how we say what we say.
The new feature helps Google to figure out what you’re saying by looking at all the variations in how you speak, from the sound of your voice to your regional accent or dialect.
If you opt into personalized recognition, we begin to associate the recordings of the words that you ask us to recognize with your Google account. We then automatically use these words to build a speech model specifically for you. This speech model enables us to deliver greater recognition accuracy. Although subtle, accuracy improvements begin fairly quickly and will build over time.
So if you like to drop your “T”s like a New Englander saying the word “mountain” (go ahead, try it out if you’re from up that way) or the words “pin” and “pen” sound nearly the same (I’m looking at you, Texas), Google can work to figure that out and offer more accurate search results. We asked Google how this worked and a spokesperson offered this tidbit of information:
Speech recognition is based on statistical modeling. To recognize spoken words, we compare the input speech to a statistical model of the language and try to find the closest match – the system’s best guess at what the user said. The statistical model is huge – it must cover all of the fundamental sounds of the language (phonemes), all of the words, and all of the different ways that the words can be strung together in the spoken language. Furthermore, it must capture all of the variations among users that happen when a language is spoken, for example all of the different dialects and accents and individual differences in the sound of the voice (e.g., male vs. female, young vs. old). Knowing what you said in the past allows us to build specialized models that are designed to match your voice and your words. Over time, this allows us to improve the speech recognition accuracy for you.
According to the announcement, the feature is completely opt-in and, upon first use, offeres the ability to tweak your privacy settings. It will be made available for English in the U.S., but already plans are in the works for other languages and countries. It is available for Android 2.2 or higher and is in the latest version of the Voice Search app in the Android Market. And, of course, you can scan this QR code to download the latest version of Voice Search: