Sung-won Park
waveform coder, LSF, wavelet transform
A low rate speech waveform coding using the wavelet transform and line spectral frequencies (LSF) is described in this paper. A speech signal is passed through an LSF analysis filter for removal of redundancy. The output of the filter known as the residual error signal is passed through a three-level wavelet transform filter bank. Only the approximation is quantized and the details are replaced by the Gaussian noise with corresponding standard deviation. Unlike vocoders whose reconstructed signal usually does not look like the original signal, the reconstructed signal of this coder still has a close resemblance to the original. As there is no substantial alteration in the waveform, the compressed speech can still be used for mental status examination.
Important Links:
Go Back
