Speech-to-speech translation concept (via U-STAR)
While it is not exactly Star Trek’s universal translator, U-STAR’s soon-to-be-released translator app looks to be pretty impressive. Developed by the Universal Speech Translation Advanced Research Consortium, the Speech to Speech Translation app (called VoiceTra4U-M) uses the ITU-T standardization protocols F.745 and H.625 that lets users speak to others in a multitude of languages such as English, Japanese and Hindi. According to U-STAR, up to 5 users can converse on a single device by face-to-face or remotely over the phone either by audio (with a slight delay) or through text for the hearing impaired.
The translation for each language is done remotely through U-STAR institute servers located in 23 different countries around the globe; meaning for those who speak Chinese, a dedicated server responsible for only that language provides speech recognition, machine translation, and acoustic synthesis for that language alone (this is the reason for the slight delay). The application was initially designed as a way for tourists to communicate to people in foreign countries where it has an 80 to 90% accuracy of translation. This is because the software, developed using Julius (speech recognition decoder) and Sphinx 4 (open-source speech recognition software), was able to learn certain words and phrases that were associated with tourist activities such as hotel/restaurant locations and directions.
Obviously using words and phrases not associated with tourist activities makes those percentages drop, so be careful at what you say otherwise describing your pet cat may have an adverse translation. This is to be expected, as the app is still in its testing phase and will continue to be so, even through the Olympic Games, until approximately March 31, 2013.
Cabe
