Low Rate Speech Coding Research-based Speech Recognition And Synthesis

In modern communication systems, voice is the most fundamental andimportant communication mode and will become the primary means for the futurehuman computer interaction. Considering the transmission cost and efficiency, thephysical channel and storage space it takes, people always hope that under thepremise of high voice quality, as much as possible to suppress the transmission ofvoice coding rate. Therefore, the voice is usually transmitted after being compressedto bits stream. This voice information compression process is called speech coding.Seen by the Shannon theory, data compression must be performed in a certaincompression limit, and these current source coding is already close to the limit. Tosacrifice the complexity of the algorithm to infinite approaching the Shannon limithas become meaningless. If we considered from the perspective of the sink (receiver),we can make content-based compression of the voice information from thetransmitting side according to the demand of the sink end, which removal of a largenumber of non-content information and greatly improve the voice signalcompression efficiency. The main contents and innovative contributions of thisdissertation are as follows:1. Thesis on the status of voice communications technology research andanalysis of very low bit rate speech coding applications and significance. Detailedanalysis and simulation are made in the key technology areas of speech recognition,speech synthesis and pretreatment and pointed out the deficiency of the present.2. Three basic characteristic parameters of endpoint detection short-timeaverage zero crossing rate, short-term energy and spectral entropy are analyzedseparately in this treatise. Based on these three impact factors，the EndpointDetection Algorithm of the Short-term EZSE (Energy-zero Spectral Entropy) isproposed.3. Trying to find a low-rate speech coding based sink related. Its theoretical basisis: the amount of information of the voice contents is always less than the voice datainformation. Based on the speech coding technology, in this paper we propose a new method encoding the speech signal by the bionic pattern recognition technique.Using this new encoding method, the text message can be received after usingbiomimetic pattern recognition of the speakers voice. And the information related tothe speakers individual characteristics can be obtained by “comparison” between thespeakers voice and the standard voice which corresponding to the text message.Then the text message and individual characteristics information are encoded beforetransmission. The transmission can be obtained a very low coding rate (<80bit/s).Atthe receiving end, using speech synthesis technology of the text information and theindividual characteristic information is converted to voice output, thereby forming acomplete voice transmission process.This paper studies is mainly used in underwater acoustic communication andmilitary communications. Because of its relatively slow communication rate, even ifthey can satisfy the real-time voice communication based on speech recognition, butfor the land, sea and air three-dimensional communication interconnect is far fromenough. Accelerate the development of our country in the field of technologicalprogress is a very urgent task.