Perceptual experiments on enhanced and slowed down speech sentences for second language acquisition

This paper investigates the perception of speech signals that have been enhanced and slowed down selectively, with the view of improving oral comprehension for second language acquisition. Our modifications are applied on a small number of acoustic cues, i.e. bursts of unvoiced stops, unvoiced fricative noises and rapid spectral transition regions. Bursts and frication noises were amplified, and spectral transitions were amplified and slowed down. We exploit energy and spectral criteria to localize bursts and frication noises, and spectral variation function to spot rapid transitions. The perception experiment involved students who learn French as a foreign language. The subjects were asked to fill in gaps in incomplete transcriptions of 50 French sentences. The average identification rate increases from 72% up to 81% when the enhancement is applied alone, and up to 86% when the two modifications are applied simultaneously. The strengths of our approach are the robustness of acoustic cue detection and the fully automatic strategy.