Speech recognition techniques have been developed dramatically in recent years. Nevertheless, errors caused by environmental noise are still a serious problem in recognition. Employing algorithms to detect and follow the motion of lips have been widely used to improve the performance of speech recognition algorithms. This paper presents a novel technique to recognize vowels. Lip features extracted by using a combined method are used as input parameters to a neural network system for recognition. Accuracy of the proposed method is verified by using it to recognize 6 main Farsi vowels.