VOICE IDENTIFICATION UNDER THE CONDITIONS OF LIMITING THE SPEECH SIGNAL BY SPECTRUM AND AMPLITUDE,

Abstract

The results of an experimental investigation of the human (male) voice identification by natural, clipped, and constant-level speech are reported. The experiments were similar to conventional articulation tests (6 speakers, 6 or 7 listeners), the task being identification of voices, not words. Russian vowels 'a,' 'e,' 'i,' 'o,' and a few words were selected as test sounds. The voices were transmitted via an electronic system comprising amplifier 1, clipper 2, differential clipper 3, constant-level-speech device 4, band-pass filter 5. These conclusions are reported: (1) Lower frequencies up to 700 cps are not important in voice identification; their exclusion practically does not affect the identification of natural speech and lowers the identification by only 3% in the case of clipped speech; (2) The frequencies beyond 4000 cps contribute only 6% in the identification; (3) The 700-4000 cps band ensures a high enough percentage (about 90) of correct voice identification on the above vowels; (4) Amplitude clipping of the signal derivative ensures greater phonetic information in the clipped speech than the amplitude clipping of the speech signal proper; (5) The average percentage of correct identification with words is higher than that obtained with vowels.

Document Details

Document Type
Technical Report
Publication Date
Jul 16, 1968
Accession Number
AD0682425

Entities

People

  • A. G. Kakauridze
  • G. S. Ramishvili

Organizations

  • National Air and Space Intelligence Center

Tags

DTIC Thesaurus Topics

  • Amplifiers
  • Amplitude
  • Frequency
  • Identification
  • Spectra

Readers

  • Mathematics or Statistics
  • Radar Systems Engineering.
  • Speech Processing/Speech Recognition.

Technology Areas

  • Microelectronics