The neural mechanisms used by the auditory system to isolate and comprehend a single speaker out of a background of other speakers are currently unknown. Young, normal-hearing individuals are capable of understanding a single speaker in complex auditory environments, while individuals with hearing loss and even some older individuals without hearing loss struggle in the same settings. A first step in ultimately designing technology to restore the ability to follow a single speaker in a crowded room is to understand how the normal auditory system achieves this feat. Human speech is composed of components that span a broad frequency range. The peripheral auditory system separates the incoming signal by spectral frequency, yet the perception of a single speaker is that of a single entity. This percept defines an auditory stream: a group of sound features that appear to originate from a single source. Pioneering work over several decades has led to a basic understanding of the psychological and physiological mechanisms of streaming simple sounds, arriving at the population separation model: the perception of unique sound sources occurs when neural responses to each source are separated into unique neural populations. For simple sounds, population separation may be inherent in the frequency separation of sources, or computed in the peripheral auditory system. However, how separation is achieved for more complex sounds with overlapping frequency content is unknown. One proposed mechanism is temporal coherence ? the idea that coincident fluctuations in the amplitude of distinct features become bound together into a single perceived stream. While the effects of temporal coherence on population separation have been tested with synthetic, periodic stimuli, these results are not likely to generalize to those of natural sounds, which contain complex temporal envelopes. Therefore, we will examine auditory streaming using naturalistic speech-like stimuli: two distinct harmonic complex tones that are modulated by temporal envelopes drawn from human speech. When modulated with different envelopes, these stimuli are temporally incoherent and evoke a strong percept of two separate streams; when modulated with identical envelopes they are coherent and are perceived as a single stream. We will collect single-neuron responses to these stimuli in marmoset auditory cortex to assess the impact of temporal coherence (and streaming) on neural encoding. We will then engage marmosets in a task requiring streaming of these sounds and quantify the effects of directed attention on neural encoding. These results will be useful in refining models of stream segregation, laying the groundwork for controlled studies using fully natural sounds.
How the human auditory system isolates and comprehends a single speaker out of a noisy background of other speakers is not fully understood. This has limited our ability to design hearing aids and other assistive devices to improve auditory communication in such scenarios. The proposed study will investigate how the brain isolates a single speaker by analyzing neural responses to naturalistic sounds in an animal model.
