LFP based classification of vocalizations in free-behaving zebra finch

D. E. BROWN, JR; E. M. ARNEODO; S. CHEN; T. GENTNER; V. GILJA

San Diego

Conferencia; SfN 2018; 2018

Society for Neuroscience

Songbirds, like humans, are one of the few species capable of learned vocal behavior, making them an attractive animal model for studying vocal learning. Understanding the neurobiological principles and mechanism that support vocal learning in songbirds, can yield useful insight into understanding human speech perception and production, and aid in the longstanding goal to develop a human speech prosthesis. With this goal in mind, we present a discrete neural decoder that predicts the vocalizations produced by an awake freely-behaving zebra finch, a species of songbird, based on local field potentials recorded in the sensorimotor telencephalic region HVC (used as a proper noun), shown in previous research to be involved with the production and timing of song.Using band power within multiple frequency bins of the local field potential (LFP) as a feature, and a linear discriminant analysis (LDA) classifier, we identified a separability in neural space for four distinct syllables and the introductory notes in a birds-own-song, and contemporaneous silence, recorded during free vocal activity. 
We computed syllable classification performance, using 4-fold cross-validation and a grid search over parameters of the LFP bins namely window length and window offset, achieving a peak syllable classification accuracy of 33±2% (mean +/- s.e.m; chance level is 16.7%). In general, classification performance increases as with window length shrinks. As we expected, better classification performance was observed when window onset close to the stimulus onset. We tested the classifier further in a series manner using a 4.5-second snippet of free vocal behavior.Our results demonstrate that the syllables of a zebra finch?s song can be ?decoded? from local field potentials sampled in HVC. Notebaly, the model described here does not utilize the temporal structure present in either the bird?s vocal behavior or the recorded neural activity. Future work will explore how consistent these dynamics are across subjects and how they correspond to volitional motor control.