Distinguishing spectral and temporal properties of speech using an information-theoretic approach

The spectro-temporal coding of Danish consonants was investigated using an information-theoretic analysis. Listeners identified eleven consonants spoken in CV[l] context. In each condition, only a portion of the original spectrum was played. Center frequencies of 750, 1500 and 3000 Hz, were presented individually and in combination with each other.

The modulation spectrum of each band was low-pass filtered at 24, 12, 6 and 3 Hz. Confusion matrices of the consonant-identification data were computed, and from these the amount of information transmitted for the phonetic features – voicing, manner and place of articulation – was calculated. From these analyses we conclude that: (1) Decoding place-of-articulation information requires significant cross-spectral integration (2) Place of articulation depends on modulations above 6 Hz, and is crucial for consonant recognition (3) Voicing requires modulations between 3 and 6 Hz. (4) Manner depends on modulations greater than 12 Hz.