الفهرس يوجد فقط 14 صفحة متاحة للعرض العام
 |  | from | 109 |  |  |
| | | from | 109 | | |
المستخلص
This thesis deals with die speech recognition problem. Many researches have developed and im¬proved different methods for speech recognition. These methods cover one of the following rec¬ognition steps:
-Ways of capturing the speech signal.
-Analyzing the captured speech to divide it into locally stationary time frames.
-Extracting useful features from each time frame.
-Assigning phonemes to time frames and thus assigning words to a sequence of frames.
’ • •
In this approach for speech recognition, the following is discussed:
- Perceptual difficulties thai make speech recognition problem a difficult one.
- The most efficieni and widely used methods for capturing speech signals, feature extraction
and matching these features to a previously stored word database.
- Software and hardware implementation of a simple stand alone speech recognizer using
ATMEL microcontrollers for signal processing and control. This recognizer can capture a word,
divide it into time frames, filter the signal with a HP, BP and LP filters, extract three dimension¬
al feature vectors for each frame by counting the zero crossing in the three filter outputs for
each frame and finally compare these feature vectors against the word database.
Although ,being simple, our stand alone recognizer is built on a modular design basis and each module represents z step of the speech recognition steps .This permits easy modification of each module to introduce a more complicated clustering, feature extraction and classification meth¬ods as discussed in chapter 2.
The chapters of this thesis are organized as follows:
Chapter 1 General introduction on speech recognition process and various fields of appli-
cations.
Chapter 2 Discusses the main steps of the speech recognition process and the most widely and efficiently methods used to perform these steps.