Sound feature detection using leaky integrate-and-fire neurons

We present a neurally inspired technique for detecting onsets in sound. The outputs from a cochlea-like filter are spike coded, in a way similar to the auditory nerve. These AN-like spikes are presented to leaky integrate-and-fire (LIF) neurons through a depressing synapse. The spike outputs from these are then processed by another layer of LIF neurons. Onsets are detected with essentially zero latency. We present results from the TIMIT database.