The object of this paper is to present and compare two different systems for isolated-digit speaker-independent speech recognition, one based in Continuous-densities HMM and the other in a Time-Delay Neural Network. The HMM system passed several optimization phases from a system with 10 Mel frequency cepstrum coefficients (MFCC) plus energy, to a system with 10 MFCC plus their 10 incremental parameters, plus energy and delta-energy. Optimization of the TDNN system has been made based on shifting some input frames for each activation of the neurons in the first layer instead of one as used in other experiments.
Bibliographic reference. Ferreiros, J. / Castro, A. / Pardo, J. M. (1991): "Comparison between two different approaches in speaker - independent isolated digit recognition", In EUROSPEECH-1991, 999-1002.