Figure 1.
The diagram presents the Training Phase and the Testing Phase. In the Training Phase, Clean Speech and Noise pass through Amplitude Normalize. Clean Speech enters the Speech Model to produce Reconstructed Speech. Noise enters Noise Model 0 to produce Reconstructed Noise. In Extension, Extra Noise passes to Noise Model N to produce Reconstructed Extra Noise. In the Testing Phase, Noisy input goes to the Speech Quality Evaluation. A decision compares Score versus Threshold. If False, a Reverse Process runs Speech Model and Noise Models with Speech Enhancement Guidance in a loop. If True, the output is Enhanced, marking End of Loop.The architecture of proposed EN-AZS method