Speech Waveform Encoding Algorithm Research Based On K-L Transform

Speech is the most direct and convenient way to human communication. While speech signal carries a large amount of voice messages, it also includes a large number of redundant information. Speech may be contaminated by noise in the process of generation and transmission. From the perspective of effective information in speech, background noise also can be regarded as redundant. To reduce the distortion with the most capability,it is the main subject to use the current multimedia technology and communication technology to realize the efficient compression by eliminating the redundancy in the speech signal. In this thesis, the author bases on thorough redundancy principle of K-L transform, and presents a coding algorithm of compressing speech waveform and denoising.K-L transform can be better to remove redundant information for signal, so signal can achieve good compression. The core of the K-L transform is to extract the orthogonal operator A, which is obtained from the signal being transformed. When the signal is a deterministic signal, operator A is fixed, when the signal is a random signal, operator A changes with the signal. Speech is a short smooth random signal, a frame of speech is smooth, therefore, orthogonal operator A of a frame of speech is fixed, but different frames of speech have their own orthogonal operator.Covariance matrix is constructed by speech frame vectors; eigenvalues corresponding to eigenvectors are got by decomposing eigenvalues of covariance matrix, orthogonal matrix is constructed by the eigenvectors; the transform coefficient vector is got by orthogonal transform of frame vectors with orthogonal matrix; obtaining the cut-off threshold M by calculating and reconstructing the new matrix by M; the transform coefficient vector is used as the inverse transformation to get the enhancement speech signal with reconstruction matrix. The enhanced speech is extracted and transmitted to the decoder, which reconstructs speech signal with interpolation technique. Under different SNR simulation experiments on different speech, and compared with the DCT encoding, The results show that the algorithm has high decoding speech, what’s more, good adaptive enhancement of speech is realized when this compression algorithm achieves 4 times.