Speaker independent word recognition using continuous matching of parameters in time-spectral form based on statistical measure