Studies Using Spectrographic Pictures from New Speech-Analysis Instrumentation

Abstract

In studies of the acoustic properties of speech elements, an advanced spectrum-analyzing system has been developed to display large quantities of information generated and recorded in real time. The speech signals are frequency-analyzed by passage through a large bank of filters producing outputs which are sampled at rates high enough to fully preserve the information. The frequency scanning is performed by a unique and flexible process of interpolation that clearly defines and accurately locates spectral energy concentrations. Thus, features such as the formants of vowels are consistently displayed by the analyzer, eliminating the need for additional calculations to determine their frequency locations. The spectral signals are recorded on photographic strip films to provide pictures which can be enlarged as needed for convenience in use. A variety of picture formats are available, including the familiar power spectrum vs time with intensity modulation to show the amplitude patterns. Accurate amplitude detail is provided by spectral cross sections taken at rates that produce complete, clear time histories of the changing speech relations. This paper presents speech pictures and data obtained by investigations using the new instrumentation. Results are presented showing variations in the acoustic characteristics of phonemes for different syllables and speakers.