Vocal communication is a critical component of the social behavior of many species, including humans.
Communication sound (vocalizations or ‘calls’) recognition is a computationally challenging problem
because calls are produced with immense variability across individuals and contexts. Our previous
research showed that call categorization, a first step in call recognition, can be accomplished by detecting
informative features of intermediate complexity in calls. Such feature selectivity likely arises in the
superficial layers of the primary auditory cortex (A1). Whether this feature detection strategy is conserved
across vocal mammals, and how neural feature selectivity develops with learning, remain unknown. This
proposal aims to answer these questions by developing a cross-species (rodents vs. primates) and
cross-level (behavioral, neural, computational) approach. In Aim 1, we will first determine whether the
selective encoding of intermediate complexity features, which we have shown to be a successful strategy
for rodent (guinea pig, GP) call categorization, is also adopted in non-human primates (marmoset
monkeys, MM). Then, using high channel count probe recordings from A1 of GPs and MMs, we will
construct network models to determine how neural receptive fields that are selective for call features can
be assembled from frequency-tuned inputs. We will compare the two species’ network models to derive
general principles of the neural circuits underlying call categorization. In Aim 2, we will train adult GPs and
MMs to categorize calls from the other species (heterospecific calls). We will determine whether
feature-selective responses to heterospecific calls emerge post-training in GPs and MMs during behavior
and the network connectivity underlying this new selectivity. Finally, we will quantify the similarities in
representation and functional network connectivity between learned and possibly innate sound categories.
These data will be used to develop a unified model for call categorization that is generalizable across
species and contexts. The proposed activities will provide insight into the neural encoding of
communication sounds, bridging behavioral, algorithmic, and mechanistic levels. The impact of this work
will be maximized through sharing of data in standardized formats, and rigorous, transparent model
validation.
