Parallel system design for time-delay neural networks

摘要

The authors develop a parallel structure for the time-delay neuralnnetwork used in some speech recognition applications. The effectivenessnof the design is illustrated by: (1) extracting a window computing modelnfrom the time-delay neural systems; (2) building its pipelinednarchitecture with parallel or serial processing stages; and (3) applyingnthis parallel window computing to some typical speech recognitionnsystems. An analysis of the complexity of the proposed design shows angreatly reduced complexity while maintaining a high throughputnrate