Accuracy of spectrographic analysis of rapidly varying formants

It has been suggested that the lack of acoustic invariance found so far in stop consonants may be attributed to the inadequacy of the traditionally used analysis methods. While the stop consonants essentially contain rapidly varying spectral information, the traditional analysis tools, such as the spectrogram, are quasi-stationary, i.e., it is assumed that the signal is stationary within a specific analysis window. In this paper the accuracy of the analysis of rapidly varying formants using spectrographic techniques is explored. Analysis of synthetic sinusoidal and speech-like signals, as well as real speech signals, shows that, when a short (e.g., 6 ms) analysis window is used, as in the traditional wide-band spectrogram, rapidly varying formants are accurately represented. A long-window representation, on the other hand, may give rise to misinterpretations, such as flattening-off of a formant and staircase-like formant tracks. It is concluded, however, that a careful interpretation of the long-window representation may still yield accurate results.