Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Neural Directional Filtering Using a Compact Microphone Array

Weilong Huang, Srikanth Raj Chetupalli, Mhd Modar Halimeh, Oliver Thiergart, Emanuël A. P. Habets

Abstract

Beamforming with desired directivity patterns using compact microphone arrays is essential in many audio applications. Directivity patterns achievable using traditional beamformers depend on the number of microphones and the array aperture. Generally, their effectiveness degrades for compact arrays. To overcome these limitations, we propose a neural directional filtering (NDF) approach that leverages deep neural networks to enable sound capture with a predefined directivity pattern. The NDF computes a single-channel complex mask from the microphone array signals, which is then applied to a reference microphone to produce an output that approximates a virtual directional microphone with the desired directivity pattern. We introduce training strategies and propose data-dependent metrics to evaluate the directivity pattern and directivity factor. We show that the proposed method: i) achieves a frequency-invariant directivity pattern even above the spatial aliasing frequency, ii) can approximate diverse and higher-order patterns, iii) can steer the pattern in different directions, and iv) generalizes to unseen conditions. Lastly, experimental comparisons demonstrate superior performance over conventional beamforming and parametric approaches.

Neural Directional Filtering Using a Compact Microphone Array

Abstract

Beamforming with desired directivity patterns using compact microphone arrays is essential in many audio applications. Directivity patterns achievable using traditional beamformers depend on the number of microphones and the array aperture. Generally, their effectiveness degrades for compact arrays. To overcome these limitations, we propose a neural directional filtering (NDF) approach that leverages deep neural networks to enable sound capture with a predefined directivity pattern. The NDF computes a single-channel complex mask from the microphone array signals, which is then applied to a reference microphone to produce an output that approximates a virtual directional microphone with the desired directivity pattern. We introduce training strategies and propose data-dependent metrics to evaluate the directivity pattern and directivity factor. We show that the proposed method: i) achieves a frequency-invariant directivity pattern even above the spatial aliasing frequency, ii) can approximate diverse and higher-order patterns, iii) can steer the pattern in different directions, and iv) generalizes to unseen conditions. Lastly, experimental comparisons demonstrate superior performance over conventional beamforming and parametric approaches.

Paper Structure

This paper contains 39 sections, 19 equations, 17 figures, 6 tables.

Table of Contents

  1. INTRODUCTION
  2. Problem Formulation
  3. Proposed Method
  4. DNN Architecture
  5. Loss Function
  6. Training Strategy
  7. Training simulation for anechoic environment
  8. Training simulation for reverberant environment
  9. Static or steerable
  10. Mini-batch sampling
  11. Performance Measures
  12. Directivity Pattern
  13. Directivity Factor
  14. Signal Estimation Quality
  15. Experimental Setup
  16. ...and 24 more sections

Figures (17)

  • Figure 1: Three directivity pattern examples on the $x$-$y$ plane. Steering direction $\theta_\textrm{s} = 0$ is used for the illustration.
  • Figure 2: DNN architecture for neural directional filtering: FT-JNF tesch_insights for static steering direction; The proposed FiLM-JNF for continuous steering direction.
  • Figure 3: (a): Optimization objective of the LS beamformer. (b): Achieved pattern by the LS beamformer with a minimum white noise gain constraint of $-15$.
  • Figure 4: Estimated power patterns regarding the $3^{\textrm{rd}}$-order DMA and $6^{\textrm{th}}$-order DMA pattern. The diameter of the array is 3 cm. The gray area in polar plots represents the standard deviation of the estimate.
  • Figure 5: Bandpass analysis of the NDF models to study the frequency processing mechanisms.
  • ...and 12 more figures