An approach based on the continuous wavelet transform and their local maxima for the speech processing is presented. We study on some issues of wavelet functions for instantaneous frequency extraction of speech signals. From then we conclude a suitable mother wavelet for the speech processing. The local maxima, which are located on the high-energy concentration curves in the time-scale domain of the wavelet transformed images, are discussed for a number of mother wavelets. They reveal the local as well as global features of the speech signal. The formants can be observed very clearly in the coarse scale modulus maximum image, whereas the pitch periods can be extracted in the fine scale modulus maximum image of the transformation. The simulation also depicts a robust approach for the case of speech signals embedded in random noise.
展开▼