SEANet Stream Enhancement on Synaptics Torq SR200

Pre-trained causal neural speech enhancement model designed for ultra-low-latency hearing aid applications.

Architecture: Causal Stream Encoder-Decoder Audio Network (SEANet)
Native Rate: 32,000 Hz mono audio
Frame Buffer: 256 samples (8.00 ms frame)
NPU Processing Latency: 1.04 ms on Synaptics Torq SR200
Total Latency: 9.04 ms (strictly under the 10.0 ms hearing aid occlusion budget)
Repository: Australian-Future-Hearing-Initiative/seanet-torq