Fuzzy Variable Frame analysis for Speech Recognition

Recent works in machine learning has focused on models such as Support Vector Machine(SVM), Artificial Neural Network(ANN) and Long Short Term Memory (LSTM), for automatically controlling the generalization and parameterization of the optimization process. This paper presents a fuzzy interpretation frame analysis procedure using LSTM classifier for noisy speech at word level using thresholding and local maxima procedure at framing level for the recognition process. Front end MFCC procedure has been modified in the framing phase to reduce the number of noisy frames using thresholding at two level local maxima procedures. A comparative results of various classifiers like SVM with kernel function, ANN and LSTM are tabulated for recognition accuracies. A fuzzy interpretation at the framing level to calculate optimal frames has been presented in this paper. In the proposed work 20% of unwanted processing of frames is reduced that equally produces the accuracies obtained by fixed frame analysis. An investigation shows that the obtained features with LSTM decrease word error rate still by 1% as increasing the recognition accuracy from 98 to 99% . approach.