Sakoe-Chiba Banded DTW Wake-Word Spotting

Sonon avoids heavy neural network speech models by utilizing Banded Dynamic Time Warping (DTW) with DBA Barycenter Averaging.

Key Advantages

1. Few-Shot Enrollment: Requires only 1 to 2 voice samples to register a new command.
2. Computational Complexity: Prunes $O(N \cdot M)$ full-matrix search to $O(N \cdot R)$ along the Sakoe-Chiba diagonal corridor ($R \approx 10$).
3. Throughput: Processes $> 2,500,000$ audio samples/sec per thread.
4. Offline & Edge Determinism: Constant memory footprint with zero cloud latency.

Enroll your own voice directly in the Playground Register Tab.