The words that a discrete speech recognition engine resolves are its vocabulary. A vocabulary consists of one or more templates for each word in the vocabulary. In speech recognition, a template is a pattern that can be used to identify or match a speaker’s pronunciation of a word. When a speech recognition engine tries to resolve a word, it compares the audio input stream to its templates until it finds a match or determines no match is available. The engine notifies the application when a speech command is recognized.
The following table shows templates supported by Windows CE for the Auto PC.
Template |
Description |
| Speaker-independent | Combines multiple pronunciations of a word or phrase, and requires no user training.
The speaker-independent templates for the vocabulary used for speech control of an Auto PC bundled applications is stored in ROM. You can install additional templates for your application in RAM. The Microsoft® Windows® CE Platform SDK for Auto PCs contains speaker-independent templates you can use in your application. You can also supply your own templates. Although speaker-independent templates are accurate for 95% of users, they may not be accurate when used by some users. |
| Speaker-dependent | Uses a single word or phrase pronunciation. Word pronunciations are trained by the application user. Training must be completed before the speech command is used. Call the IVCmdUserWord::Train function to set up user training in your application.
Although training can take some time, speaker-dependent templates can be useful if the user has a non-native accent or other speech patterns that do not reliably match the speaker-independent templates. |