The perceptual flow of phonetic feature processing

How does the brain process spoken language? It is our thesis that word intelligibility and consonant identification are insufficient by themselves to model how the speech signal is decoded - a finer-grained approach is required. In this study, listeners identified 11 different Danish consonants spoken in a Consonant + Vowel + [l] environment.

Each syllable was processed so that only a portion of the original audio spectrum was present. Three-quarter-octave bands of speech, centered at 750, 1500, and 3000 Hz, were presented individually and in combination with each other. The conditional, posterior probabilities associated with phonetic-feature decoding were computed from confusion matrices in order to deduce the temporal flow of phonetic processing.

Decoding the feature, Manner-of-Articulation, depends on accurate decoding of the feature Voicing (but not vice-versa), and decoding Place-of-Articulation requires precise decoding of Manner (but not the converse). From these data, we conclude that Voicing is processed prior to Manner-of-Articulation, and that Manner is decoded prior to Place-of-Articulation.

Voicing and Manner cues are often correctly decoded in conditions where Place is not. This asymmetric pattern of feature decoding may provide extra-segmental information of utility for speech processing, particularly in adverse listening conditions.