Study of the use of linear predictor and related methods of speech analysis for measuring vocal-tract area functions
A. Maynard Engebretson, N. Rao Vemula · The Journal of the Acoustical Society of America · 1974
The linear predictor (LP) and partial correlation (PARCOR) methods of speech analysis have been applied to sounds produced by acoustic horns with inside cross-sectional shapes comparable to the human vocal tract. Our experience with human speech suggests that the area functions obtained from both methods are variable and depend strongly on the type of speech sounds, the sampling rate, the method of windowing and compensation of the signal, and other factors. To evaluate these methods in a controlled situation, the horns are driven with a known input volume velocity. The output volume velocity is measured, sampled at a 20-kHz rate, and analyzed with a computer program using either an LP of PARCOR algorithm. The computed area function and transfer function are compared with the known geometry and measured frequency response of the horn. Implications of these results on the use of LP and related methods for studying vocal-tract area functions in hearing and nonhearing subjects are discussed.