Low complexity decomposition for the characteristic waveform of speech signal

For efficient coding of speech, it is desirable to separate the slowly and rapidly evolving spectral components to take advantage of their different perceptual qualities. Existing decomposition methods are too inflexible to model transient changes in the speech signals, require high delay or produce a large parameter set that is not scalable to low rates. We present a low complexity decomposition method, based on SVD, applied to waveform interpolation (WI) coding. This scheme reduces the computational complexity of the common SVD method in WI by exploiting the properties of human auditory perception to lower the dimensions of the decomposition matrix. This method requires only a single frame of speech and overcomes the substantial delay problems. The quantization solution involves the use of vector quantization on the separately decomposed singular matrices, U, V, and the diagonal matrix of singular values, S. The quality of the reconstructed speech can be varied according to the scalable decomposition and the bit rate available.