A perceptual study of source coding of Fourier phase and amplitude of the linear predictive coding residual of vowel sounds
Changxue Ma, Douglas D. O’Shaughnessy · The Journal of the Acoustical Society of America · 1994
A practical question in a Fourier transform coding of speech signals is to what accuracy their amplitude and phase spectra have to be represented without perceptible distortions. In this paper a concern is with the audibility of quantization noise signals which are produced by quantizing the amplitude and phase spectra of vowel sounds. Experiments show that the detection of the noise targets with maskers of a low fundamental frequency is determined mainly by the sharpest temporal resolution of the auditory system in the high-frequency region. For maskers of a high fundamental frequency the detection is determined mainly by the sharpest spectral resolution in the low-frequency region. Noise targets with global random phase and amplitude are relatively more difficult to detect than those produced by a local randomization. Local random phase noise targets are generally more detectable than those produced by local amplitude randomization. The relative importance of phase and magnitude spectra in the Fourier transform coding is strongly dependent on the fundamental frequency of the vowel sounds and the window size used in the short-time Fourier analysis.