Development of gender and race recognition system using speech and recognition by using frequency spectrum

• Main Author: Ng Siew Fong
• Other Authors: Sazali Yaakob, Prof. Dr. (Advisor)
• Format: Learning Object
• Language: English
• Published: School of Mechatronics Engineering 2009
• Subjects: Recognition system > Design and construction; Speech recognition; Speech processing systems; Automatic speech recognition; Speech perception; Linear prediction coefficient
• Online Access: http://dspace.unimap.edu.my/xmlui/handle/123456789/4600

In this thesis, the development of an algorithm and system that is able to recognize gender and races by using the speech frequency spectrum is presented. Some of the features extracted are the formant frequency and fundamental frequency of speech signals. The formant frequencies are obtained by finding a set of predictor coefficient that minimizes the mean square error over a short segment of speech waveform. Whereas the fundamental frequency is estimated by finding a peak in the auto-correlation function with a corresponding delay. Results obtained using this approach to identify the gender by using formant frequency and fundamental frequency has proven to be practically applicable. The Back-Propagation Neural Network has been chosen for the classification purposes. These features will be fed into the neural network for training until it is able to outputs the appropriate gender and races. The speech recognition application is implemented using the Digital Signal Processor (DSP). The well trained network parameters were applied in the DSP by the new architecture support features that facilitate the development of efficient high level languages. Hence, C code was chosen to read a set of input features, as the weights are being adjusted and weights sum output from input features, the corresponding results are finally displayed on the Liquid Crystal Display (LCD) through DSP.