Why The Tellme Acquisition By Microsoft Matters
In case you missed the news, Microsoft has agreed to acquire Tellme for a rumoured c. $900M. Why did Microsoft pay almost a billion dollars for the company?
Well, there are a number of components that go into the valuation. However, the thing that’s really exciting about Tellme is what they’re doing in the area of mobile search. Last December, I wrote about what some top VCs wanted to invest in:
The final mobile idea is from Danny Rimer at Index Ventures. He wants to spend $2M for this idea… He wants a mobile search system, with a user interface that’s designed for mobiles (rather than the standard ports of PC systems that Google and the like do). I think it’s a good idea. The speech recognition part could be useful as a part of the interface. I’ve done a fair amount of work on speech recognition in the past, and I think this kind of mobile search application could play to the strengths of speech recognition technology. However, I suspect we’re probably a year (or more) away from being ready to include speech recognition as a core part of mobile apps. I agree with him about the problem of data entry on mobile phones too, which makes many current apps unusable; and I do think that effective mobile search could be transforming if you can get the right kind of integration of things like shopping search, driving directions and even special offers.
Well, here’s the deal. I was wrong about us being about year or more away from being able to include speech recognition in mobile applications. Why? Because Tellme have already done it; and they did it by completely side-stepping the key technical problem that was the reason I said we couldn’t do this yet. That technical problem is that speech recognition on mobile devices doesn’t yet work very well, and speech recognition mobile APIs are a way off.
How did Tellme side-step this problem? Simple. They record your voice, and upload the digital recording to their servers via the network… and their servers do the speech recognition. So, they have a system that works today. And the timing for this kind of solution might be just about perfect: all-you-can-eat data plans are coming down in price (and a typical voice search is < 4KB in size for the upload); and mobile operators are rolling out high-speed networks.
If Microsoft and Tellme, together, execute well on mobile search, that would be worth many billions of dollars.
Check out the great interview with one of Tellme’s founders on the CalacanisCast, and great piece featuring the Tellme CEO over on CNET.
Asam Bashir wrote:
If you need to press buttons, then record a voice, then press a button to send to servers, it’s not going to work. It has to be quicker then being able to key in a query, for that you still need some voice activation on the phone side. Which is still possible, but it’s going to need an increadibly slick GUI and system to deal with that.
Also, even on the server side, how is any voice recognition going to deal with variabilty in language and reginal accents? This has always been the problem with voice recognition, if you’re a US English speaker you’re ok, but have you tried those recognition programs with even an English accent? You end up having to talk with a silly American accent, so it becomes a little embarrising to use if you’re in the presence of other people. Until you have recognition in different languages, non-English speakers will have major problems. Imagine a French speaker, using the recognition in English, don’t see any software recognition technology being able to handle that.
Posted 21 Mar 2007 at 6:16 pm ¶
simon wrote:
A few coments on your comment:
Re: The UI. I haven’t tried the TellMe beta. However, there’s no reason for this to multiple clicks. It can be “push to talk” and that’s it. So, you push and hold the button, say your query, e.g. “pizza, San Francisco”; then release the button. That’s it. Then you get your results.
It could be pretty slick - the typical voice sample for this is 2KB-4KB in size i.e. fast to upload.
Re: the state of speech recognition. I don’t know how well TellMe’s systems work. Yes, there are problems with speech recognition if people don’t talk clearly; but if they do, the systems can handle different accents pretty well. As I say, I don’t know how well TellMe’s systems work - but they have built their main business sucessfully around speech recognition. So, it must be at least “good enough”. And, you could believe that by centrally processing millions of voice samples, they could have done something innovative here in terms of dealing with different voices. They might not have done - but it is *possible* they have…
Posted 21 Mar 2007 at 7:39 pm ¶