Sound localization

Sound localization#

This experiment uses only the submillisecond difference between left- and right-ear arrival times to drive a spiking direction readout. Exact delay taps, event communication, and batched independent state make timing semantics the core of the implementation.

Prompt#

Build a spiking network that tells whether a sound came from the left or right using only the tiny difference in when it reaches the two ears.

Agent decision path#

  1. Classify the system as a point-neuron auditory circuit.

  2. Route neurons to brainpy-state, spike delivery to brainevent, and physical timing to brainunit.

  3. Select the Jeffress delay-line principle with two auditory relays and a coincidence bank.

  4. Round physical interaural delays once to an explicit 0.05 ms time grid.

  5. Deliver binary spikes through fixed-fan-out projections to left and right readouts.

  6. Give every interaural delay independent mapped membrane, refractory, and delay state.

  7. Advance all lanes through one state-aware vmap2 step inside for_loop.

  8. Verify delay impulse convention, dimensional bounds, and direction labels with three tests.

Result#

Every tested negative interaural time difference is classified as RIGHT, every positive difference as LEFT, and simultaneous arrival as CENTER. All three focused tests pass.

The selected run did not generate a plot. This structured result preserves the committed classification evidence without inventing a graph.

With-BrainX skill/Without skill comparison#

With BrainX skill

Source captured · benchmark pending

Add measured rollout throughput across the same delay grid, including compile policy and all timing-semantic tests.

Without BrainX skill

Matched run not collected

Use the same 0.05 ms grid, +/-0.6 ms bounds, and direction convention before comparing speed or correctness.