Fig. A.

Signal-processing stages in the S-AMPH model. (a) Original acoustic waveform of the spoken sentence “Mary Mary quite contrary”. (b) In the S-AMPH model, the original speech signal is first filtered into 5 frequency bands, and the Hilbert envelope is computed for each frequency band. (c) A 3-tier AM hierarchy is then extracted from the envelopes of each frequency band. The resulting ‘Stress’ (0.9–2.5 Hz), ‘Syllable’ (2.5–12 Hz) and ‘Phoneme’ (12–40 Hz) AMs are shown overlaid in different colours. These correspond to prosodic stress patterns, syllable patterns and phoneme patterns respectively. This results in a 5 (frequency band) × 3 (AM hierarchy) spectro-temporal representation of the speech amplitude envelope.