•  
  •  
 

Corresponding Author

Tolba, Ahmed

Subject Area

Electrical Engineering

Article Type

Original Study

Abstract

A natural man - machine communication with word processors requires the integration of speech recognition techniques. Speech recognition task is a difficult problem because of its temporal nature and the possibility of the presence of noise. Therefore, it is difficult to extract significant features by using the traditional techniques. Artificial neural networks introduce simple and fast learning techniques for this problem. This paper introduces the application of a three layer neural network for converting spoken Arabic words into Arabic text. We describe a supervised learning method which is based on the well known back propagation technique. The designed network proves itself to be speaker independent and noise immune word spotting guarantees invariance under translation in time. A user friendly word processor should nave intelligent interfaces such as the speaker interface and the hand writing interface. In this work we introduce the first interface. It is expected that neural network based speech-to-text transcription systems should have a significant effect on office

Download

Share

COinS