Speech Classification for Single Word by Using the Feature of Front Aspiration

• Main Authors: Te-Cang Hsiung, 熊德昌
• Other Authors: Yu-Kumg Chen
• Format: Others
• Language: zh-TW
• Published: 2005
• Online Access: http://ndltd.ncl.edu.tw/handle/20511504769044308267

碩士 === 華梵大學 === 電子工程學系碩士班 === 93 === By using the colors of pixels and the shapes of image, the character of sound can be found in a two-dimensional image of spectrogram. Thus, the image of spectrogram is applied in the speech recognition extensively.A new real-time front of aspiration recognition algorithm of isolated speech is proposed in this paper. It can recognize the aspiration effectively for the isolated speech of different languages. With the energy method, the proposed algorithm finds out the compendious end-point of isolated speech first. Then, by using the binary searching on the sampling pixels of the column section of the image of spectrogram, the proposed algorithm can derive the more accurate initial end-point of the isolated speech based on its compendious end-point. Therefore, it can be used in the real-time front of aspiration recognition of isolated speech. Since the proposed algorithm just uses a few pixels in the image of spectrogram, it will reduce the amount of the operations and it is very suitable for applying to the mobile equipment. Experiments are carried out for Chinese digits and English letters to demonstrate the computational advantage of the proposed method.