THE EFFECT OF FREQUENCY FILTERING ON CONSONANT RECOGNITION

JAMES DAVID LAMBERT, Purdue University

Abstract

Lambert, James David. Ph.D., Purdue University, August 1954. The Effect of Frequency Filtering on Consonant Recognition. Major Professor: M. D. Steer.An investigation was designed to study the effects of frequency filtering upon the recognition of certain consonant sounds when those sounds are uninfluenced by other speech sounds. Specifically, the study was designed to answer the following questions:1. Will the elimination of certain frequencies below2000 cps result in lowered recognition scores for the consonant sounds of English?2. Will the elimination of certain frequencies above 2000 cps result in lowered recognition scores for the consonant sounds of English?3. Will such filtering affect different sounds differently, i.e. will stop consonants be affected differently from fricative consonants; will there be differences within the stop and fricative groups?4. Where the removal of frequencies from the spectrum results in a reduced ability to recognize the consonant sounds, what substitutions are most likely to occur?Six stop consonants p, b, t, d, k, g 7 and nine fricative consonants (f, v, s, 2, ts, dz, S, 0.27 were combined with the neutral vowel Int so as to form fifteen nonsense syllables of the CV type. These fifteen syllables were recorded by a young male adult who had normal voice quality and articulation. Using these fifteen syllables and by means of a process of dubbing, cutting, and splicing tapes, four lists of seventy-five randomly presented syllables were constructed. These lists were then read to forty young college adults with normal speech and hearing who recorded the consonants they heard. Ten conditions of filtering were employed--high-pass 1000, 1200, 1500, 1700, and 2000 cps and low-pass 1000, 1200, 1500, 1700, and 2000 cps. The results were analyzed for the effects of the various filter levels, for differences in their effect upon the sub-sets of consonants, and for their effect upon specific sounds. The substitutions made for sounds mis-identified were inspected for pattern of substitution related to filter effect.Within the limits set by the experimental procedure and the techniques of analysis used, the following conclusions are drawn:1. When all the frequencies above 2000 cps are present, the loss of frequencies below that level has little, if any, effect upon the recognition of the sounds < k g, f, s, z, JtJ*d_{3} , J.2. When all the frequencies below 1500 cps are present, the loss of frequencies above that level has little, if any, effect upon the recognition of the sounds angle K_{i} s, J, tJ, d_{3}*J3. The range of frequencies of primary importance in the recognition of the sounds J lies above 1500 ops. k, B, f, s, z, S, ts, d_{3} ,4. The recognition of the stop consonants J ^ k * J and [87 and of the fricative consonants [ts], [ d x\ ] and initial [v] which might be described as "stop-like", is materially affected by the loss of frequencies below1500 cps.5. The loss of frequencies above 2000 cps results in a greater proportion of errors among the voiced sounds than among the voiceless sounds but shows a tendency for sounds of like quality to be substituted for the voiced and the voiceless sounds.6. Loss of the frequencies below 2000 ops shows no preponderance of errors among the sounds of either class but does indicate a greater tendency for a voiced sound to be substituted regardless of the type of sound missed.

Degree

Ph.D.

Subject Area

Theater

COinS