Google's Nexus One takes the #### out of voice recognition

 

Gallery

Posted on 25 Jan 2010 at 16:01

A Reuters correspondent has uncovered one of the Google Nexus One's less publicised features: its innovative speech-to-text function censors swear words, replacing them with hash (#) marks.

The Nexus One is the first Android phone to offer speech-to-text, which uses Google's speech recognition service. Users of the Nexus one can use voice input for any text field; you simply tap on a microphone icon on the keyboard and dictate emails, texts or documents.

However the technology is still in its infancy and isn't as accurate as Google would like, so it's added a word filter as a safety valve to avoid offending prudish customers. "We filter potentially offensive or inappropriate results because we want to avoid situations whereby we might misrecognize a spoken query and return profanity when, in fact, the user said something completely innocent," Google told Reuters.

The speech recognition technology is typical of the way Google operates. The service is primarily used for its Google Voice telephony service, where it transcribes voicemail messages into emails, but it's also used in other services.

For example, Google uses it on YouTube to automatically generate captions for videos, and it's used in Google GOOG-411 directory enquiries service in the US. There's also a Google Labs experiment to transcribe text from video, but it seems it was created specifically to monitor speeches made during last year's US elections and has since been abandoned.

As more and more people use these services, Google will build up a huge repository of voice data which Google engineers can use to improve the speech recognition algorithms, in the same way that it is able to improve search suggestions based on what other people have searched for. We'd like to offer a suggestion to Google: just fit a microphone into Gordon Ramsey's kitchen and your speech recognition algorithms will soon learn which words need blanking.