Use of Two Nonacoustic Measures in Computer Recognition of Spoken Digits
William A. Hillix · The Journal of the Acoustical Society of America · 1963
Instrumentation was developed for obtaining a measure related to lip and jaw movement and a measure related to the velocity of air pulses in front of the lips during speech. These measures were used as inputs to a computer program designed to categorize decimal digits after “learning” on a sample of the digits. Waveform recordings of the measures were also presented to human observers for visual categorization. For the first measure alone, sets of digits produced by one speaker were categorized with 71% accuracy by the computer and 78% accuracy by the observers. The observers did somewhat better with another speaker. However, performance of the observers fall to about 40% when categorization of one speaker's digits was attempted using another speaker's digits as samples. These data, however, were obtained under unfavorable circumstances. When the air-velocity measure was added, the observers' accuracy increased to about 95%, and the computer's performance rose to 81% under one set of circumstances and 100% under another. It is felt that the results to date have been promising enough to indicate the potential usefulness of such measures in understanding and recognizing speech.