ANALYSIS OF SPEECH RECOGNITION CRITERIA,
Abstract
A method and apparatus is proposed for analyzing the criteria for speech recognition. The first part of the work consists of studying the Russian vowel sounds (a), (o), (i), (u), and (e). Oscillograms were made for each of these sounds. The spectrum of each sound was examined. The spectral analysis of the vowels was carried out on the ASChKh-1 spectral analyzer. The controlled analysis of the frequency spectrum was carried out on the Ural-1 computer. On the basis of a preliminary analysis of the oscillograms, an apparatus was set up for discerning vowel sounds recorded by one informant. The same procedure was followed for the consonants. These were divided into three groups: nasals, fricatives, and occlusives. Seven informants took part in this study, four men and three women. Preliminary analysis of the oscillograms showed the feasibility of studying the amplitude-frequency-time characteristics of these sounds. An apparatus was set up for conducting the above study. The sound under study is recorded on a tape, amplified, and passed through a system of filters with variable frequency tuning, it is then demodulated and quantized according to levels. The equipment used in this experiment is not objective in recognizing sounds. It is useful only for recognizing the sounds of a single speaker. (Author)
Document Details
- Document Type
- Technical Report
- Publication Date
- Apr 25, 1968
- Accession Number
- AD0680571
Entities
People
- E. T. Semenova
Organizations
- National Air and Space Intelligence Center