Neurally driven synthesis of learned, complex vocalizations.

Area of Science:

• Neuroscience
• Bioacoustics
• Bioengineering

Background:

• Brain-machine interfaces (BMIs) are crucial for motor function restoration and studying motor skills.
• Existing BMIs for limb prosthetics utilize nonhuman primates, but speech prostheses lack suitable animal models and advanced neural interfaces.
• Songbirds offer a compelling model for complex learned vocal behavior, sharing similarities with human speech and vocal motor control.

Purpose of the Study:

• To develop a novel vocal synthesizer for birdsong using neural population activity.
• To establish a new animal model for studying learned vocalizations and advancing BMI technology.
• To demonstrate the direct synthesis of complex natural behaviors from neural data.

Main Methods:

• Recorded neural population activity from the HVC region of songbirds using electrode arrays.
• Mapped neural activity onto low-dimensional compressed song representations using real-time computational methods.
• Utilized a generative biomechanical model of the syrinx as the target for neural mapping to synthesize vocalizations.

Main Results:

• Successfully demonstrated a real-time vocal synthesizer for birdsong.
• Synthesized vocalizations that accurately matched the individual bird's song.
• Provided proof of concept for direct synthesis of complex behaviors from neural activity.

Conclusions:

• This study establishes songbirds as a viable model for vocal BMI research.
• The developed vocal synthesizer demonstrates the potential for creating species-specific prosthetic vocalizations.
• The findings suggest broader applications for synthesizing complex behaviors from neural data across different species.