It’s amazing the stuff you run across when you have time to just surf the Web, or at least catch up on your e-mail over a holiday. My friend and colleague Ray Horak sent me a link to an article that ultimately led me to this posting on Bell Labs’ site. What the researchers there have done is to build a prototype system that could allow a cell-phone user to “speak” via mouthing the desired words and be understood at the other end – but without making a sound.
The principle behind this breakthrough is pretty simple – just use reflected ultrasound to analyze the vocal tract during the mouthing of speech, and then use a synthesizer to generate sound that is the equivalent of speech. I’ve not tired this and I suspect that (a) it’s not going to sound better than other speech synthesizers, and (b) it’s going to suffer from the same high error rate as other recognition technologies, such as speech-to-text systems and “OCR”, which isn’t all that optical anymore. This whole area is, after all, part of the field of artificial intelligence, and we’ve yet to build a computer that’s as good at basic pattern-matching skills as your average five-year-old. It’s a hard problem.
But if Bell Labs can get this stuff to work, and it can be commercialized in an inexpensive, portable, low-power implementation, then two of my key objections to speech as a primary communications vehicle – disturbance caused to others nearby, and a lack of security – go away. That’s why this may very well be, indeed, a breakthrough.
