A New Time-Domain Analysis of Human Speech and Other Complex Waveforms.

Abstract

The purpose of this research is to explore the usefulness of a new time-domain analysis of complex waveforms, especially with respect to human speech. Essentially three separate investigations are presented, with the last two predicated on the results of the first: (1) Cycle-based time-domain parameters were extracted from the speech waveforms of many hundreds of utterances, and were then subjected to extensive scrutiny, both by hand and by machine. (2) Based solely on time-domain phenomena found in the previous study, the authors wrote an automatic segmentation program for continuous speech. (3) They examined the time-domain acoustic characteristics of 228 allophones of fricatives and stop consonants, for each of three speakers (2 males, 1 female). Finally, they present a personal view of the synergism inherent in the utilization of these time-domain techniques with the traditional frequency- domain techniques. In addition, suggestions are presented for applying these generalizable time-domain techniques to other complex waveforms, especially amenable to such analysis. Specific examples are drawn from music (e.g. violin) and animal (e.g. bou-bou shrike) vocalizations.

Open PDF

Document Details

Document Type
Technical Report
Publication Date
May 01, 1975
Accession Number
ADA013583

Entities

People

  • Janet Maciver Baker

Organizations

  • Carnegie Mellon University

Tags

Communities of Interest

  • Biomedical

DTIC Thesaurus Topics

  • Acoustic Phenomena
  • Acoustic Signals
  • Amplitude Modulation
  • Artificial Intelligence
  • Automated Speech Recognition
  • Birds
  • Computer Science
  • Computers
  • Databases
  • Electrical Engineering
  • Frequency
  • Frequency Domain
  • Identification
  • Information Processing
  • Recognition
  • Speech Analysis
  • Waveforms

Readers

  • Speech Processing/Speech Recognition.
  • Theoretical Analysis.