Pulse Coupled Neural Network (pcnn) In The Spectrogram-based Speaker Recognition

Spectrogram contains speech information rally in time and frequency domain. It has been being studied since it came out. Some outcome was done, but not good. In 1990s, after they had studied mammalian ocular pallium, Eckhorn's team founded PCNN (Pulse Coupled Neural Network) through simulating the pulse output phenomenon in mammalian ocular pallium.Since PCNN came out, it has been being used in image processing. Proposed in this paper, a method of image compression using PCNN and run-length encoding is one special illustration of PCNN using in image processing. Yet spectrogram is the image expression of speech. So using PCNN in spectrogram processing is reasonable. We can apply the methods of image processing to speech processing based on spectrogram.When extracting the feature of spectrogram by PCNN, people usually put spectrogram into PCNN and get 50 data points about pulse number from output of PCNN after PCNN runs 50 times. But this feature only stands for some information of the spectrogram, not the whole. A new method is given in this paper, which extracts feature from spectrogram using PCNN in different way. The feature extracted by new method shows more information of spectrogram than the former one.The main works of this paper are here:1. Introducing the principle and structure of Pulse Coupled Neural Network (PCNN) and simplified PCNN. And listing out the parameters of PCNN.2. Presenting a new method of image compression using PCNN and run-length encoding to show how PCNN is used in image processing.3. Doing experiment about speaker verification and speaker identification with feature of 50 data points' output from PCNN. And analyzing the results of experiment.4. Selecting parameters of PCNN by Genetic Algorithm.5. Analyzing pulse position images of PCNN in different time. Proposing a new feature extracting method from pulse position images. Doing experiments of speaker verification and speaker identification with new feature. The results of experiment show that the new feature is better.