Televirtual’s UK Media Lab has just given birth to METman, a virtual weather reporter/forecaster, whose entire performance is generated automatically from a few lines of text-based data issued as a meteorological summary, and accompanied by a weather map update.
METman doesn’t even need a script
In the final application, raw field-entry data, in the form of facts and figures will be fed into the system, which automatically draws from a lexicon of appropriate phrases, to produce a narrative description of events past, and those to come. This text is then fed into METvoice the first ever artificial voice or TTS (Text-to-Speech) engine, to be custom-built to broadcast standards.
The human vocal model for METvoice was Televirtual boss and founder, Tim Child, a broadcaster and former TV newsman.
Modern speech engines are created by recording up to 30 hours of dictated speech, but by capturing Tim’s speaking patterns as well as words and phrases, Televirtual were able to ‘fine-tune’ the new engine to a performance level unheard of to date..
The new speech engine has further powerful features. Operating as part of Televirtual’s award-winning RAP animation system, METvoice features a powerful XML-style mark up language stream, triggering lip-synch animations, and controlling and dictating the 3d animated METman’s
moods, expressions, gestures, and screen positions.
Whilst the new voice is still being improved upon, the early results are impressive. ‘Unless you were aware or suspected it, you would not normally be able to detect METman’s vocal performance as anything other than the real thing,’ said Tim.
Further ‘broadcast’ voices are now planned, and the breakthrough is by no means limited to weather forecasting. Gaming channels and Quiz TV variants could also employ the system to operate virtual presenters in virtual sets, at a fraction of the cost of conventional presentation methods.
But the big market for such synthetic voice and character creation applications is probably in the Home of the near Future.
Installed in domestic television set top boxes (STBs), 3d ‘announcers’ will be able to present a personalised information service tailored to individual requirements.
Such ‘homecasters’ would be able to advise on TV viewing schedules, read the news and weather on demand, and trawl the internet on request.
In disability scenarios, they will be able to read incoming emails to the blind, and could ‘sign’ to the deaf or provide lip-readable augmented information to the hard of hearing, whilst the speech-impaired might use them to make phone calls.
Kaynak : 