The principle objective of this research is to study and introduce the sixth generation computer-Neural Networks as an affective and excellent tool to solve the unsolved or weakly solved problems. In this research we propose a new Neural Networks model - The Smart model and build an Arabic speech synthesizer with unlimited vocabulary of two speakers; male and female . Both the Smart model and the Arabic speech synthesizer are used in designing and implementing the proposed system - Arabic NETtalk . The Arabic NETtalk system takes any non-diacritized Arabic script and process it by using the Smart model to generate the fully-diacritized script and the narrow phonetic transcriptions that is processed by the Arabic speech synthesizer to produce an understandable intelligent Arabic speech (male or female). Many experiments and tests have been made to improve the system performance. The Arabic NETtalk system has three phases; training, speaking and dreaming. After training the system on 100 words selected from one text randomly, it can pronounce any one of these words with 100% accuracy and any other word with 65% accuracy (in this case, the system will automatically train itself to pronounce such words with 100% accuracy during the dreaming phase). The Arabic NETtalk system was implemented on an IBM/PC compatible under Arabic DOS 5.0 or higher, with a sound blaster support using Turbo C 2.0 .