Real-time information reduction in digital sound spectograms of speech

A new method has been employed in order to reduce the loss of relevant information while converting digital sound spectrograms into their two-level representations. The method involves a threshold which is adaptively adjusted in time as well as in frequency. When compared with other reported attempts, the resulting binary spectrograms yield much more reliable indications of formants, formant transitions, turbulences, bursts, etc. Moreover, a feasible digital hardware solution of a simplified matrix processor is shown to be sufficient for the real-time computation of the binary spectrograms.