Talk (to your) Radio
Nice news from VoiceBox Technologies (just around the corner in Kirkland, WA), who are currently on the front page of CNET News announcing a deal with XM Satellite Radio to "let drivers issue voice commands to search through XM Satellite Radio's 160 channels of music, talk radio and other stations and to find personalized information such as stock prices, traffic and weather."
I hope we'll see a lot more of these kinds of personalized information services made available through speech within the next year or two. For an in-car content provider like XM it's logical: the subscription model already enables the personalization of content, and enabling a hands-free speech interface is a compelling UI (assuming that the claim that VoiceBox's technology "worked accurately in noisy environments" is true enough). What would be even more interesting would be speech access to a broader range of portals - with a user profile, there's data that's useful in so many ways for speech and dialog processing, the possibilities get exciting very quickly. Personalized dialogs will work so much better because in time (a) they'll be more accurate, because any stochastic components can be adapted to individual behaviour, (b) they'll be more useful because they'll let users do their common tasks more quickly, and (c) they'll appear to be more intelligent, because they 'know' about a user's habits. Let me know if you're doing things in this space or if you've seen anything interesting.
A side note: XM's a big name, hence the CNET interest, but it's interesting that CNET's 'related items' failed to add anything to do with speech recognition or in-car telematics - the Big Picture diagram (usually graphically informative) and Related Stories tab show nothing from this field, even though the VoiceBox technology is at the core of the story. There could be a number of reasons for this. 1. The associative filtering must be directly keyword- or keyphrase-based, and VoiceBox story vocabulary doesn't really hit on any main buzzwords (instead of using phrases like "speech recognition", and "natural language", it says "voice control", "voice commands" and "free-form conversational language". And there's not a single mention of "automotive" or "telematics"). 2. XM is a big enough news magnet in itself that the closest associations are other XM-related stories that crowd out the more interestings technology angles ("XM subscriber base tops 5m", "Is Stern worth his millions"...) 3. The story also added some news on 5.1 surround sound broadcasting technology, and this added extra terms from the audio domain to crowd out the speech telematics even more. Anyway, another example of the weakness of doing information retrieval (including search) on a purely lexical basis.
Update (7 Jan 2006): Rob Chambers joined the queue for the demo in the Lexus at CES. Read his report here!
