Event-driven spectrotemporal feature extraction and classification using a silicon cochlea model

Area of Science:

• Digital Signal Processing
• Auditory Neuroscience
• Neuromorphic Engineering

Background:

• Traditional auditory processing systems face limitations in real-time, low-power computation.
• Event-based systems offer a promising alternative for efficient sensory data processing.
• Implementing complex cochlear models on hardware requires efficient digital architectures.

Purpose of the Study:

• To present a reconfigurable digital implementation of an event-based binaural cochlear system on an FPGA.
• To introduce an event-driven SpectroTemporal Receptive Field (STRF) Feature Extraction using Adaptive Selection Thresholds (FEAST).
• To evaluate the system's performance against existing auditory processing techniques.

Main Methods:

• Digital implementation of Cascade of Asymmetric Resonators with Fast Acting Compression (CAR-FAC) cochlea models.
• Integration of leaky integrate-and-fire (LIF) neurons for event-based signal processing.
• Development and application of the event-driven FEAST algorithm for feature extraction.

Main Results:

• The reconfigurable FPGA implementation successfully processes binaural auditory signals in an event-driven manner.
• The proposed FEAST method effectively extracts spectro-temporal features using adaptive thresholds.
• Performance evaluation on the TIDIGITS benchmark shows competitive results compared to state-of-the-art approaches.

Conclusions:

• The developed FPGA-based binaural cochlear system provides an efficient and reconfigurable platform for event-based auditory processing.
• The novel event-driven FEAST feature extraction method shows significant potential for auditory signal analysis.
• This work contributes to the advancement of neuromorphic auditory sensing and processing.