HHT Based Speech Eomotion Recognition Technique Research

Speech emotion recognition is an important part of the field of affectivecomputing and more and more researchers and institutions engaged in this field inrecent years. Usually we have to perform the framing operation before applyingFFT method to the voice fragment. In this article, the Hilbert-Huang transformmethod is used in our research. The HHT method is very useful tool to handle thenon-linear and non-stationary signals and the entire sentence could be treated as aunit to be processed. This article has focused on several following works.1. By using the Hilbert-Huang Transform method, a new set of long-termfeatures based on Hilbert marginal spectrum has been proposed. Compared withtraditional long-term features based on FFT method, these features gain greatadvantage in recognizing the shorter sentences in the emotional corpus. We alsogive a detailed description about the impacting factors of the experimental results.2. Based on the characteristics of the Empirical Mode Decomposition, thesetwo methods extracting the structural features and component features are proposedand we also study deeply the differences of the intrinsic mode functions with thesame order among the different emotional sub-corpuses. And then the details ofscheme solving the problem is designed and accomplished.3. On one hand, we study the influences of each intrinsic mode function to allthose emotional sub-corpuses, on the other hand,we analyze the contributions fromall those IMFs under the same emotion. The resulting data may somewhat reflectthe emotional information distribution in the frequency domain and this may give ussome ideas to make up our mind how emotion expresses in the speech.4. The measure turns out to be very powerful which combined the thresholdmethod and the iteration-time limited method together to solve the time-consumingproblem of the empirical mode decomposition algorithm.