Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

A Novel Deep Learning Framework for Efficient Multichannel Acoustic Feedback Control

Yuan-Kuei Wu, Juan Azcarreta, Kashyap Patel, Buye Xu, Jung-Suk Lee, Sanha Lee, Ashutosh Pandey

TL;DR

Acoustic feedback in multichannel devices causes howling and limits gain; traditional DSP struggles with highly correlated feedback signals. The paper proposes a Convolutional Recurrent Network (CRN) to fuse spatial and temporal processing for efficient multichannel acoustic feedback control, trained via In-a-Loop Training, Teacher Forcing, or a Hybrid MCWF strategy. The framework demonstrates robust suppression of howling and high PESQ scores across varying gains and delays, with an architecture of 684K parameters and 82 MACs per second, suitable for edge deployment. This method advances AFC for devices like hearing aids and smart glasses, enabling higher gains with lower latency and practical real-world performance.

Abstract

This study presents a deep-learning framework for controlling multichannel acoustic feedback in audio devices. Traditional digital signal processing methods struggle with convergence when dealing with highly correlated noise such as feedback. We introduce a Convolutional Recurrent Network that efficiently combines spatial and temporal processing, significantly enhancing speech enhancement capabilities with lower computational demands. Our approach utilizes three training methods: In-a-Loop Training, Teacher Forcing, and a Hybrid strategy with a Multichannel Wiener Filter, optimizing performance in complex acoustic environments. This scalable framework offers a robust solution for real-world applications, making significant advances in Acoustic Feedback Control technology.

A Novel Deep Learning Framework for Efficient Multichannel Acoustic Feedback Control

TL;DR

Acoustic feedback in multichannel devices causes howling and limits gain; traditional DSP struggles with highly correlated feedback signals. The paper proposes a Convolutional Recurrent Network (CRN) to fuse spatial and temporal processing for efficient multichannel acoustic feedback control, trained via In-a-Loop Training, Teacher Forcing, or a Hybrid MCWF strategy. The framework demonstrates robust suppression of howling and high PESQ scores across varying gains and delays, with an architecture of 684K parameters and 82 MACs per second, suitable for edge deployment. This method advances AFC for devices like hearing aids and smart glasses, enabling higher gains with lower latency and practical real-world performance.

Abstract

This study presents a deep-learning framework for controlling multichannel acoustic feedback in audio devices. Traditional digital signal processing methods struggle with convergence when dealing with highly correlated noise such as feedback. We introduce a Convolutional Recurrent Network that efficiently combines spatial and temporal processing, significantly enhancing speech enhancement capabilities with lower computational demands. Our approach utilizes three training methods: In-a-Loop Training, Teacher Forcing, and a Hybrid strategy with a Multichannel Wiener Filter, optimizing performance in complex acoustic environments. This scalable framework offers a robust solution for real-world applications, making significant advances in Acoustic Feedback Control technology.

Paper Structure

This paper contains 12 sections, 11 equations, 3 figures.

Table of Contents

  1. Introduction
  2. Methods
  3. Acoustic Feedback System
  4. Default System
  5. Feedback-Controlled System (MISO Model)
  6. Training Methods
  7. In-a-Loop Training
  8. Teacher Forcing Training
  9. Hybrid with Multichannel Wiener Filter
  10. Experimental Setup
  11. Results
  12. Conclusion

Figures (3)

  • Figure 1: Block diagrams illustrating the three training approaches for acoustic feedback control: (a) In-a-Loop Training, (b) Teacher Forcing Training, and (c) Hybrid with Multichannel Wiener Filter (MCWF).
  • Figure 2: Comprehensive evaluation of acoustic feedback control models showing PTPR, SNR, and PESQ across varying gain levels. Demonstrate the models' capabilities in managing howling, enhancing signal clarity, and maintaining speech quality under different operational conditions.
  • Figure 3: Performance evaluation of the CRN model under varying time delays.