Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Dynamic nsNet2: Efficient Deep Noise Suppression with Early Exiting

Riccardo Miccini, Alaa Zniber, Clément Laroche, Tobias Piechowiak, Martin Schoeberl, Luca Pezzarossa, Ouassim Karrakchou, Jens Sparsø, Mounir Ghogho

TL;DR

This work presents an early-exiting model based on nsNet2 that provides several levels of accuracy and resource savings by halting computations at different stages and adapt the original architecture by splitting the information flow to take into account the injected dynamism.

Abstract

Although deep learning has made strides in the field of deep noise suppression, leveraging deep architectures on resource-constrained devices still proved challenging. Therefore, we present an early-exiting model based on nsNet2 that provides several levels of accuracy and resource savings by halting computations at different stages. Moreover, we adapt the original architecture by splitting the information flow to take into account the injected dynamism. We show the trade-offs between performance and computational complexity based on established metrics.

Dynamic nsNet2: Efficient Deep Noise Suppression with Early Exiting

TL;DR

This work presents an early-exiting model based on nsNet2 that provides several levels of accuracy and resource savings by halting computations at different stages and adapt the original architecture by splitting the information flow to take into account the injected dynamism.

Abstract

Although deep learning has made strides in the field of deep noise suppression, leveraging deep architectures on resource-constrained devices still proved challenging. Therefore, we present an early-exiting model based on nsNet2 that provides several levels of accuracy and resource savings by halting computations at different stages. Moreover, we adapt the original architecture by splitting the information flow to take into account the injected dynamism. We show the trade-offs between performance and computational complexity based on established metrics.
Paper Structure (13 sections, 4 equations, 6 figures, 2 tables)

This paper contains 13 sections, 4 equations, 6 figures, 2 tables.

Table of Contents

  1. Introduction
  2. Deep noise suppression
  3. Dynamic Architecture
  4. Split Layers
  5. Training
  6. Layer-wise Training
  7. Joint Training
  8. Experimental setup
  9. Dataset
  10. Evaluation Metrics
  11. Model Configurations
  12. Results
  13. Conclusion

Figures (6)

  • Figure 1: nsNet2 architecture with exit stages (dotted lines show an example of full inference path)
  • Figure 2: Different styles of split layer adaptations
  • Figure 3: Boxplot of quality metrics at different exit stages.
  • Figure 4: Comparison of performance/efficiency trade-off at different exit stages, relative to 1 second of data.
  • Figure 5: Comparison of example noisy input (in color), suppression masks (in grayscale) at different exits, and baseline.
  • ...and 1 more figures