Epoch Extraction from Pathological Children Speech Using Single Pole Filtering Approach

The instant of significant excitation of the vocal tract system is referred to the epoch of the speech signal. The presence of high pitch and aperiodicity are the major challenges for the epoch extraction from the speech of pathological children. In this work, impulse-like characteristics of epochs derived from single pole filter based time-frequency representation are exploited to propose an epoch extraction algorithm for the pathological children speech. The sharp transitions present in the single pole filtered envelope at the epochs are enhanced using multi-scale product computation. Further, the combined evidence derived from the multi-scale product of the filtered envelopes at different frequencies is used to locate the epochs. The proposed algorithm is evaluated over the Saarbruecken Voice Database containing pathological children speech and simultaneously recorded electroglottographic signals. The proposed method showed better identification accuracy for pathological children speech when compared to state-of-the-art techniques.