Using the subband-dft in speech cepstrum analysis and echo detection

The subband-DFT (SB-DFT) [1],[2],[3] is examined for new applications in cepstrum analysis and echo detection. In many speech processing applications, the DFT-based cepstrum is used to determine the mode of excitation of the model (voiced or unvoiced) and, for voiced speech, the pitch period [4]. In this work the approximated subband-DFT (SB-DFT) is used instead of the full-band DFT (or FFT) to determine both the cepstrum (real cepstrum) and the complex cepstrum. It is shown that using the approximate SB-DFT in speech applications does not degrade the information contained in the cepstrum while the computational complexity is highly reduced. The approximated complex cepstrum is used in the problem of echo detection. The adaptive technique added to the SB-DFT in [5] is included to adaptively detect the echo for different input frequency bands.