Optimal frame rate analysis for speech recognition
S.J. Young, David Rainton · 1990
The authors examine the process of sampling the underlying time-frequency distribution and show that the degree of smoothing applied to each estimate involves a trade-off between minimising the bias and the variance. This leads naturally to the use of a time variant smoothing window and a mean square error criterion for determining the optimum window spread function. They also present some preliminary results from applying these ideas to the speech recognition problem.