A Learning AI orithm of Neural Networ for Spectrum Envelope ation
Hiroyuki Kamata, Nobuharu Matsumoto, Yoshihisa Ishida · 1996
We try to estimate the spectrum envelope of speech by using neural networlts. The neural networks system proposed in our study consists of two neural networks parts. The one is the network for detecting the spectrum peaks, another is the one for interpolating between the extracted peaks. The structure of the latter is derived from the Neville interpolation. The former is based on the peak picking method. We presented an operator for detecting the peaks using the neural networks. Unfortunately, the result is that the neural networks do not only the general peaks but also the local peaks and the extremely closed peaks. In this paper, we try to train the neural networks to detect only the general peaks. A leaming algorithm presented in this paper uses steepest descent method such as the back- propagation (BP) algorithm. After the learning, we show the detected peaks and the spectrum envelope when we use the leamed pattern as the inputted data. In addition, we make an experiment whether the proposed operator corresponds to the XOR boolean function or not.