摘要:
A neuro-fuzzy filter device that implements a moving-average filtering technique in which the weights for final reconstruction of the signal are calculated in a neuro-fuzzy network according to specific fuzzy rules. The fuzzy rules operate on three signal features for each input sample. The signal features are correlated to the position of the sample in the considered sample window, to the difference between a sample and the sample at the center of the window, and to the difference between a sample and the average of the samples in the window. The filter device for the analysis of a voice signal includes a bank of neuro-fuzzy filters. The signal is split into a number of sub-bands, according to wavelet theory, using a bank of analysis filters including a pair of FIR QMFs and a pair of downsamplers; each sub-band signal is filtered by a neuro-fuzzy filter, and then the various sub-bands are reconstructed by a bank of synthesis filters including a pair of upsamplers, a pair of FIR QMFs, and an adder node.
摘要:
A device for filtering electrical signals has a number of inputs arranged spatially at a distance from one another and supplying respective pluralities of input signal samples. A number of signal processing channels, each formed by a neuro-fuzzy filter, receive a respective plurality of input signal samples and generate a respective plurality of reconstructed samples. An adder receives the pluralities of reconstructed samples and adds them up, supplying a plurality of filtered signal samples. In this way, noise components are shorted. When activated by an acoustic scenario change recognition unit, a training unit calculates the weights of the neuro-fuzzy filters, optimizing them with respect to the existing noise.
摘要:
A device for filtering electrical signals has a number of inputs arranged spatially at a distance from one another and supplying respective pluralities of input signal samples. A number of signal processing channels, each formed by a neuro-fuzzy filter, receive a respective plurality of input signal samples and generate a respective plurality of reconstructed samples. An adder receives the pluralities of reconstructed samples and adds them up, supplying a plurality of filtered signal samples. In this way, noise components are shorted. When activated by an acoustic scenario change recognition unit, a training unit calculates the weights of the neuro-fuzzy filters, optimizing them with respect to the existing noise.