|
|
Speech recognition
|
|
| Speech recognition is an area of application for pattern
recognition and includes four steps: |
|
| 1. |
Digitalisation
Analog speech is turned into a digital pattern by the sound card. The
size of the sample needed depends on the length of the recording and the
quality of the analog-digital converter. |
|
| 2. |
Modulation
The digital pattern is transformed into its most characteristic features.
The algorithm must work fast and the result must be as compact as possible.
An algorithm that identifies two words as having the same characteristics
is useless. |
|
| 3. |
Matching
The program searches in a database for the word matching a given pattern.
The vocabulary consists of a table in which the significant characteristics
are associated with the corresponding words.
A word can only be recognized by the system if the database contains
a corresponding pattern. If the pattern is not in the database, the word
can be "learned" by the application. |
|
| 4. |
Output
The found word is generally saved in memory for immediate use in other
applications. |
|
|
|
|
|
|
|
|
