TravellingWave, which develops next-generation mobile input technology, has launched what it claims is the first voice-powered text prediction application for mobile phones. The companys VoicePredict software will enable mobile users to input text using their voices and by typing one or two letter key presses per word spoken. 
Additionally, VoicePredict will enable the quick insertion of emoticons, SMS-shortcuts, and contact names, and can act as an interface to social networks, including Facebook and Twitter.  VoicePredict, currently in a beta release, is available as a downloadable application from a dedicated website for Windows Mobile Smartphones. Travellingwave expects to deliver versions for iPhone and Android in early 2010.
TravellingWave will market the software, which is currently in a limited trial, to a range of customers, including handset manufacturers, mobile operators, developers of mobile operating systems, and of speech recognition applications and text prediction software. The company is also engaged in a multiyear public-private program, funded by the National Science Foundation, through Carnegie Mellon University for the control of assistive devices, such as wheelchairs.
VoicePredicts beta launch marks an industry breakthrough, says Todd Achilles, Principal of VC firm, Balius Ventures.  No one has thought to combine the two domains – speech recognition and text prediction – in the way that TravellingWave has done. VoicePredicts multi-domain approach means that speech recognition is more accurate and text prediction is faster, than if either solution functions alone. This is T9 squared. TravellingWave has successfully flipped the speech-to-text problem on its head and the result marks a new chapter in how people interact with devices.
VoicePredict is based on a new approach to speech input, wherein voice inputs and keypad inputs are tightly coupled to result in voice-powered text prediction. VoicePredict enables reliable voice and text communication that can be easily personalized, based on the users choice. Preliminary trials indicate that, on an average, VoicePredict results in a 75% reduction in the number of key-presses on a phone with a Qwerty keypad, and enables 80% faster insertion of symbols and contact names.
The value proposition of VoicePredict is that it resides on a device in the form of an orthogonal input that is always listening and significantly enhancing all other hand/finger based input modes, using a speech recognition solution that is near 100% reliable, says TravellingWave CEO, Dr. Ashwin Rao. Our underlying algorithms process background noise, extract human pitch and other acoustic features, perform acoustic-phonetic classifications, and finally dynamically integrate all of these with key-based inputs, for real-time voice prediction. The redundancy in acoustic and language information is what separates VoicePredict from all other speech-to-text and text prediction solutions.
VoicePredict offers word choices that are 95% accurate, with just one or two letter inputs; if the user desires not to speak but simply type, or if the background noise is unsuitable for speech recognition, then VoicePredict automatically falls back to keypad-based text prediction.