Coherent versus incoherent modulation filtering.
Marc Brennan, Pamela E. Souza, Les Atlas, Adam R. Greenhall · The Journal of the Acoustical Society of America · 2010
Understanding how amplitude modulations are used for speech recognition is essential for improving speech signal processing techniques. Previous studies using vocoding or incoherent modulation filtering indicated that slow amplitude modulations are essential for speech recognition [e.g., Drullman (1994), Shannon (1995)]. Each method had limitations that constrain our understanding of how speech is decoded. These methods did not account for a product model which has the modulating signal multiply the carrier signal in speech or used modulation filtering which retained inherent carrier components. Coherent modulation filtering [C. & Atlas (2009)] is a modulation filtering method that uses a product model with more effective modulation filtering. The purpose of this study was to establish differences in speech recognition performance with coherent versus incoherent modulation filtering and assess the importance of slow amplitude modulations in speech recognition while preserving the carrier signal. Subjects were adults with normal hearing. Consonant recognition scores were measured for 16 nonsense syllables at different modulation filter cutoff frequencies. Speech was presented at a conversational level in a closed-set task. Recognition improved at high modulation filter cutoff frequencies for both conditions. At low-modulation filter cutoff frequencies, performance was substantially poorer with coherent compared to incoherent modulation filtering.