| Speech communication, as one of the most direct and efective means of communica-tion, has experienced the change from analog communication to digital communication. Inthe future, people would like speech communication services to be of higher quality and lowerprice. Speech coding is an important part of speech communication technology, an excel-lent speech coder is the premise of high-quality speech communication services. Today, theneed for a variety of speech communications makes the communication channel increasinglycomplex, which leads speech coding technology development towards high-quality and lowrate. With the development of the large scale integrated circuit, the major chip manufacturershave provided a variety of high performance digital signal processing chips, for the real-timeapplication of speech coding algorithms.Sinusoidal models have played a very important role in analyzing and coding speechsignals. This class of models is based on short-time speech analysis technology, and the char-acteristic parameters of the speech signal are amplitude, phase and frequency. Low-rate speechcoding based on sinusoidal model is a hot research topic in the area of speech coding. How-ever, estimating the phase parameters is very difcult, and the improved sinusoidal model–the instantaneous amplitude model, can be a good solution to this problem. Instantaneous am-plitude model is characterized by two instantaneous amplitudes and frequency parameters, sowe can get the phase parameter by two instantaneous amplitudes and avoid the detection ofthe phase parameter.In this thesis, we build two speech codecs with diferent quantization methods, one is ascalar quantization codec operating at17.2k bit/s, the other is vector quantization codec at6kbit/s. Simulations show that the two codecs have high quality of reconstruction speech. Weimplement the instantaneous amplitude model codec on TI5509A DSP chip. Experimentsshow that the synthetic speech have a good quality. |
PDF Full Text Request