Neural coding of the temporal envelope of speech: Relation to modulation transfer functions

Continuous speech shows pronounced low-frequency modulations in its temporal envelope. Modulation frequencies near the average syllabic rate of 3-4 Hz are the most prominent. Degradations in these low-frequency modulations reduce speech intelligibility (Drullman et al., 1994; Drullman et al., 1994; Houtgast and Steeneken, 1973), while speech processed to have minimal spectral information is intelligible providing that low-frequency modulations are preserved (Shannon et al., 1995). Thus, low-frequency modulations are both necessary and almost sufficient for accurate speech reception. The ability of auditory neurons to encode amplitude modulation has been characterized using modulation transfer functions (MTFs) (reviews by Eggermont, 1993; Langner, 1992). The MTF expresses, as a function of frequency, the complex ratio (magnitude and phase) of the modulation in the neural response to the modulation in the acoustic stimulus. Previous neurophysiological studies have rarely focused on low modulation frequencies most important for speech reception, and have not explicitly related the MTFs of auditory neurons to their response to speech. In the present study, we measured both MTFs and responses to a speech utterance for single units in the auditory nerve (AN), cochlear nucleus (CN), and inferior colliculus (IC) of anesthetized cats. We also developed a functional model for predicting the neural response to speech based in part on the MTF.