Enhanced Time-Locked Decoding for Spoken Words but Not Environmental Sounds in Natural-Like Auditory Conditions

Humans are especially sensitive to speech sounds even in complex acoustic environments, but it remains unclear whether speech is tracked differently from other meaningful sounds under such conditions. Magnetoencephalography recordings combined with machine learning have revealed that dynamic time-locking of cortical activation to unfolding speech is crucial for encoding its acoustic-phonetic features. Here we investigated whether a similar mechanism for speech encoding operates during concurrent processing of speech and nonspeech sounds. Twenty participants listened to superimposed spoken words and nonspeech environmental sounds while attending to one stream at a time. Using a time-locked decoding model, we reconstructed time-varying acoustic characteristics of attended and ignored sounds from neural responses in each hemisphere. In the left hemisphere, amplitude envelopes of attended spoken words were decoded significantly better than those of attended environmental sounds at 120- to 200-ms latency between the sound and the cortical activation. No such difference between sound types emerged in the right hemisphere. Furthermore, attention significantly enhanced the decoding of speech sounds in the left hemisphere, suggesting that the observed effects reflect both bottom-up and top-down driven processes. These findings imply that particularly the left hemisphere processes speech sounds in a special, time-locked manner even under adverse auditory conditions. This mechanism may share its neural underpinnings with the previously reported time-locked tracking of speech amplitude envelope by cortical oscillatory activation or by evoked responses to acoustic edges, supporting efficient extraction of speech features in natural-like listening conditions.