الفهرس | Only 14 pages are availabe for public view |
Abstract Speech signals are the language of communication among people anywhere in the world. To transmit these signals through the channel, they need a large number of bits. Thus, they require large channel bandwidth. Therefore, speech coding and compression are considered as the solution to this problem. Speech coding has a vital role in the speech processing area. Speech coding converts the analog speech signal into compressed binary form. The goal of the conversion process is to reduce the number of bits needed for transmission. Thus, the cost is decreased. This thesis is concerned with efficient coding and compression of speech signals. In addition, the effect of decoded and decompressed speech signals on the Speaker Identification (SI) for remote access systems is studied. In this thesis, speech coding and two compression techniques are used. The applied speech coding technique is the Linear Predictive Coding (LPC) as it is the most popular technique in mobile communications. The first compression technique for speech signals depends on the decimation process for compression, and thus the original speech signal is reconstructed using inverse techniques. Inverse techniques include maximum entropy and regularized solutions. On the other hand, the second compression technique is Compressed Sensing (CS). The coding and compression techniques are compared and the performance of the recovered signal is evaluated using two metrics; the Perceptual Evaluation of Speech Quality (PESQ), Dynamic Time Warping (DTW). The results prove that the CS technique works efficiently in the absence and in the presence of noise. Also, in this thesis the effect of decoded or decompressed speech signals on the performance of SI system in the remote access scenario is investigated. For the SI system, feature vectors are captured from different discrete transforms such as Discrete Wavelet Transform (DWT), Discrete Cosine Transform (DCT) and Discrete Sine Transform (DST), in addition to the features from the time domain. Finally, a comparison between the effects of all speech communication scenarios on the SI system is presented. Simulation results prove the success of speaker identification process even in the presence of reconstruction loss and channel effect. |