Accord V6 Forum banner

1 - 1 of 1 Posts

4,681 Posts
Discussion Starter #1
This is quick version of re-organization on this topic... Thanks to Galen on making this wonderful write-up...

ps - i will try to re-organize this and many other information when i finally got time not to go out to meets.. :p

Original thread....

Basic understanding of voice recognition
To effectively use the voice recognition system it helps to understand some of the basics behind voice recognition. What follows is a hopefully brief summary of the technology available, what the navi system uses, and how to use it more effectively.

There are basically three modes of recognition:
Limited keyword list - A system using this mode can recognize a finite and usually relatively small (a few hundred words would be large) set of distinct words. It recognizes word-by-word, and has trouble if two words in its dictionary are similar. For example, it would have trouble with "wear" and "where", or "six" and "sex". No voice training is necessary, as it performs a "best match" search.

Grammar - A system using this mode can recognize more words, and recognizes based on a progressive context, defined in the form of a grammar. A grammar might look like this:
Phrase: Locate [Place]
[Place]: [Food-type] Restaurant | Movie Theater | Gas Station | ATM | (etc)
[Food-type]: Italian | Mexican | Seafood | Fast Food | (etc)
So a system with this grammar could recognize any phrase beginning with the word Locate, followed by anything in the [Place] list, or followed by anything in the [Food-type] list followed by the word Restaurant. It may be able to more easily tell the difference between words like "Eighteen" and "Eighty" based on the grammar, in the same way the street number keypad entry limits what digits you can enter. If the system is looking for a number between 1 and 20, and it hears "ay-ee", it will most likely translate that as "eighteen". This is what the Accord navi system uses.

Dictation - A system using dictation can recognize virtually any word in its tens-of-thousands-of-words dictionary, but it MUST be trained to your voice. It takes anywhere from 30 minutes to 2+ hours to properly train a dictation system, as it must get samples of your voice matched against known words and phrases, so it can pull out the individual phonemes and learn how you pronounce them.

Accord's NAVI system
The Accord navi system uses a context-based grammar. Certain commands are only usable from certain screens. There are a fairly large number of global commands, though, which slows the system down and makes it a bit less accurate. The fewer things it has to search against, the faster and more accurate it will be. To see how the grammar works, try giving it a nonsense command. For example, tell it "XM Channel bindey floor". If it can parse it (it should be able to), it will most likely turn to XM channel 94, because when it hears "XM Channel" it is expecting to then hear a number - so it will try to match whatever you say against its known set of numbers [Digit][Digit?][Digit?] or [Tens][Digit?] or [Hundreds][Tens?][Digit?].

With this in mind, if you take the time to learn its grammar - the Navi owners manual describes it but doesn't express it formally or completely - you will be better able to give it commands, by knowing what it expects to hear. Absent a full knowledge of the grammar, there are still some things you can do to improve recognition:
* Give clear separation between words. This helps it to identify individual words and map them into the grammar.
* Speak clearly, but not too loudly - volume can distort the phonemes.
* Enunciate.
* When you push the voice button, you will hear a beep. Wait for the beep to start talking, otherwise it may miss half a phoneme or two.
* As the owner's manual suggests, limit background noise.
* Limit the size of its grammar by using the touch screen or buttons to go one or two menus deep before switching to voice recognition.
1 - 1 of 1 Posts