Wednesday, October 19, 2011

"My hovercraft is full of eels"


Although built on an excellent voice recognition engine and using it's servers, Siri sometimes need a little help to understand what you are trying to say. At times it can feel like this infamous Monty Python sketch which coined this posts title.  It is unclear how much of the engine learns over time from your speech, but she purports to adapt to your speaking as you increase her use. For now, some things can help ease the learning curve between man and machine...girl. Like her carbon-based counterparts, Siri sometimes acts like she only hears what she wants to hear. Although light years from the handwriting recognition of the Apple Newton and much better than early cellphone voice dialing attempts ("call Cindy"...."now calling Mindy"), Siri however still needs some teaching or cheating to understand some of your commands.

Since many other applications use the same Nuance inspired VR engine (which was also born at SRI where Siri was also hatched before sold to Apple in 2010), some of those experiences can help make the relationship with Siri work even better.  Apps like Dragon Dictation can provide some history of the "nuances" of the VR engine and many of the commands seem to work.  Here is also a list of commands that should work and give you some ideas.

There are also some other things that can help.

Siri can be easily coaxed or tricked into thinking she is right when the situation arises.  As an example, when referring to a stored contact "Costco", Siri had a little trouble at first recognizing the name as spoken. She did not know what "cosco" was when it was enunciated.  The store was correctly spelled and saved in the Contacts file, so what is up Siri?  Well, being her literally perfect self, she was mostly right for being a bit confused. When many people pronounce the name of the large membership warehouse store, they pronounce it as "coss co", not as it is spelled, "cost co".  The solution was to either pronounce it as spelled with the "t" properly voiced, or do a little slight of tongue.

The Contacts card has a couple of helpful things which can solve this problem without having to change your improper diction.  Simply edit the Contact name and Add a field and either chose to add a Nickname which is entered as you say it, "cossco".  You can also chose to add it phonetically in the provided Phonetic field for first name.  Now she can understand your mangled use of the language and you don't need to misspell the contact name.  Siri can also be corrected on the fly by actually retyping the misinterpreted name when written back by Siri for your approval or clarification.


It also appears that steady commands with no hesitation work best, so as always with dictation systems, know what you are going to say before asking her to listen to you.  She can be a bit impatient.

No comments:

Post a Comment