Using Siri

With a new iPhone 5 on the Ver­i­zon LTE net­work, I’ve been exper­i­ment­ing with Siri for sim­ple tasks and infor­ma­tion requests. Get­ting dis­ap­point­ingly mixed results. Until I learned how to get the best results from Siri, I found myself wast­ing time try­ing to cor­rect Sir­i’s mis­takes. But when she works prop­erly, it’s delight­ful.

What I’ve learned is that my nor­mal speak­ing voice and cadence cause Siri to make too many errors. To get the best out of Siri, I’ve had to learn to slow down, speak more sim­ply, use key­word-style dic­tion, and enun­ci­ate more clearly.

I try to avoid flow­ing sen­tences or words that sound as if they run together. As a result my inter­ac­tions with Siri sound some­what robotic, but they get the job done…

As an exam­ple of cadence prob­lems, when I asked Siri to launch the web page for Ama­zon’s wine store, Siri heard “swine,” and dis­played a Google search results page with infor­ma­tion about pigs.

When I ask Siri to find some­thing that is stored on my phone (such as con­tact or cal­en­dar info), I do so using key­words that often appear on the iPhone. So I get bet­ter results when I ask Siri to “call Dad Wilson’s home phone” rather than “call Dad.” Sadly, my take­away on how best to use Siri is that it’s yet another exam­ple of humans hav­ing to learn from the tech­nol­ogy, rather than vice versa.

That said, I’ve found dic­tat­ing text mes­sages via iMes­sage (on the iPhone or my new Mac­Book Pro) to be a lovely time-saver. The con­tent of text mes­sages is usu­ally quite sim­ple, so the lack of com­plex­ity dri­ves bet­ter results with Siri.

Net net: there are occa­sional moments of delight, clear time-savers and con­ve­niences in a nar­row domain, but the over­all results are still under­whelm­ing. Apple knows that Siri is not yet ready for prime time, so Apple posi­tions Siri as a beta prod­uct.

For­tu­nately, tech­nol­ogy lead­ers like Apple and Microsoft are hard at work to find ways that will enable us to speak more nat­u­rally to or through our devices, and get the results we expect.

Microsoft: What’s Cooking in the Lab

Microsoft Research has begun demon­strat­ing a more accu­rate, real-time speech recog­ni­tion capa­bil­ity that also lever­ages machine lan­guage trans­la­tion tech­nol­ogy. Microsoft claims this speech recog­ni­tion approach dri­ves 30% fewer mis­takes than the other tech­nolo­gies on the mar­ket today. This 10-minute video shows Rick Rashid, head of Microsoft Research, demon­strat­ing this tech­nol­ogy. It’s worth watch­ing to see how far they’ve come in the lab.

First you’ll see real-time tran­scrip­tion of his speech (like next-gen closed cap­tion­ing) — amaz­ingly good, but not error-free. Then you see the tech­nol­ogy trans­lat­ing Rashid’s speech into Chi­nese Man­darin; first as writ­ten lan­guage and then spo­ken lan­guage.

What’s amaz­ing is that the spo­ken ver­sion of the machine-lan­guage trans­la­tion attempts to mimic the speak­er’s nat­ural tones and cadences. Just imag­ine the improve­ments when this evolves from the lab to a com­mer­cial qual­ity prod­uct!

I think the Microsoft tech­nol­ogy demon­stra­tion gives us a peek at what the future has in store for us. A world in which we can use our own voice to more accu­rately direct our devices to carry out spe­cific task requests, as well as one that enables us to com­mu­ni­cate with other peo­ple across lan­guage bar­ri­ers.