Synchronous Activity in the Auditory Forebrain Represents Temporal Landmarks in Vocal Communication Signals
Abstract
Precise temporal segmentation of sound is critical for identifying and interpreting complex acoustic signals such as human speech (Pisoni and Luce, 1987) and bird song (Gentner, 2008). Using the zebra finch model system, we investigated the neural coding of the songs' temporal features. We recorded neural activity from avian primary and secondary auditory cortex-like brain areas using electrode arrays in awake birds (four males and three females) exposed to conspecific songs. By applying a deep-learning approach with bidirectional long short-term memory networks (Daly, 2023), we demonstrate that not only the amplitude envelope of song stimuli but also discrete, time-locked envelope landmarks can be decoded from multichannel neuronal responses to songs. The times of envelope peak amplitude and peak slope were decoded with average accuracies of 40% and maximum accuracies of 70% at 20 ms timescales. The decoding of these landmarks enabled the segmentation of song into syllables with maximum accuracy of 90% at 10 ms timescales. Decoding accuracy for temporal events was greater in the primary auditory than in the secondary auditory area, suggesting a serial processing of the segmentation and identification operation. These findings reveal that neural entrainment to salient temporal events in vocal communication signals is a ubiquitous, specialized feature of the vertebrate primary auditory forebrain (Mesgarani et al., 2009; Heelan et al., 2019) providing a mechanism for segmenting continuous acoustic streams into behaviorally relevant units for higher-order processing.