Adding a zero-crossing count to spectral information in template-based speech recognition

Zero-crossing data can provide important feature information about an utterance which is not available in a purely spectral representation. This report describes the incorporation of zero-crossing information into the spectral representation used in a template-matching system (CICADA). An analysis of zero-crossing data for an extensive (2880 utterance, 8 talker) alpha-digit data base is described. On the basis of this analysis, a zero-crossing algorithm is proposed. The algorithm was evaluated using a confusible subset of the alpha-digit vocabulary (the "E-set"). Inclusion of zero-crossing information in the representation leads to a 10-13% reduction in error rate, depending on the spectral representation. This research was sponsored in part by the National Science Foundation, Grant MCS-7825824 and in part by the Defense Advanced Research Projects Agency (DOD), ARPA Order No. 3597, monitored by the Air Force Avionics Laboratory Under Contract F33615-78-C-1551. The views and conclusions contained in this document are those of the authors and should not be interpreted as representing the official policies, either expressed or implied, of the Defense Advanced Research Projects Agency or the US Government