Binaural LCMV Beamforming With Partial Noise Estimation

Nico Gößling, Elior Hadad, Sharon Gannot, Simon DocloIEEE/ACM Trans. ASLP 2020 · OpenReview

TL;DR

This paper introduces the Binaural Linearly Constrained Minimum Variance with Partial Noise Estimation (BLCMV-N) beamformer, a novel algorithm for hearing devices. It effectively balances reducing background noise and interfering sources while preserving the listener's spatial perception, outperforming existing methods in perceptual tests.

What Problem Does This Paper Address?

Hearing devices, especially those with a binaural setup (two microphones, one in each ear), aim to improve speech intelligibility in noisy environments. Beyond simply reducing unwanted sounds (like background babble or other speakers), a crucial goal is to maintain the "spatial impression" of the acoustic scene. This means preserving the natural cues (like the slight differences in sound arriving at each ear) that allow us to perceive where sounds are coming from and how diffuse a soundscape is.

Existing methods have limitations:

  • BMVDR (Binaural Minimum Variance Distortionless Response): Good at noise reduction and preserving the desired speech's spatial cues, but it distorts the spatial cues of interfering sources and background noise, making everything sound like it's coming from the same direction.
  • BLCMV (Binaural Linearly Constrained Minimum Variance): Can control the reduction of interfering sources while preserving their spatial cues, but it offers less noise reduction than BMVDR and doesn't allow for fine-tuning the preservation of background noise's spatial cues.
  • BMVDR-N (BMVDR with Partial Noise Estimation): Mixes BMVDR output with the original noisy signal to control background noise spatial cues, but its effect on coherent interfering sources is less predictable.

This paper aims to combine the strengths of BLCMV and BMVDR-N to achieve better overall performance.

Methods

The core contribution is the BLCMV-N beamformer. It's built upon the BLCMV framework but incorporates a "partial noise estimation" component, similar to the BMVDR-N. This is achieved by introducing two new parameters:

  • η (eta): A mixing parameter that controls how much the original noisy microphone signals are blended with the beamformed output. This is analogous to the BMVDR-N's approach for managing background noise.
  • δ (delta): An adjusted interference scaling parameter. This parameter, derived from the original BLCMV's interference scaling parameter and the new mixing parameter η, allows for fine-tuning the suppression of interfering sources.

The paper derives mathematical expressions for the BLCMV-N's performance and shows that its output can be viewed as a weighted combination of:

  1. The original noisy microphone signals.
  2. The output of a BLCMV that preserves the desired source and nulls the interfering source.
  3. The output of a BLCMV that preserves the interfering source and nulls the desired source.

This decomposition highlights how the BLCMV-N can simultaneously control interference suppression and background noise spatial cue preservation.

Datasets

The paper uses two main types of data for evaluation:

  1. Measured Anechoic Acoustic Transfer Functions (ATFs): These were obtained from behind-the-ear hearing aids mounted on a head-and-torso simulator (HATS). This data is used to validate the theoretical derivations of the algorithms.
  2. Reverberant Recordings: These were recorded in a cafeteria environment with a competing speaker and multi-talker babble noise. This provides a more realistic evaluation scenario.

No new datasets were introduced; existing, publicly available (or described in cited works) datasets were utilized.

Key Insights and Ideas

  • Bridging the Gap: The BLCMV-N successfully merges the benefits of BLCMV (interference control and preservation) and BMVDR-N (background noise spatial cue control) into a single framework.
  • Tunable Spatial Perception: The introduction of the mixing parameter η and the adjusted interference scaling parameter δ provides fine-grained control over both noise reduction and the preservation of spatial cues for interfering sources and background noise.
  • Decomposition Reveals Functionality: The mathematical decomposition of the BLCMV-N into weighted sub-beamformers clearly illustrates how it achieves its performance by combining different signal processing strategies.
  • Trade-offs are Quantifiable: The paper provides theoretical and experimental evidence showing how noise reduction performance is traded off against binaural cue preservation for background noise, and how interference reduction is controlled by δ.
  • Perceptual Validation is Crucial: Objective measures are important, but the perceptual listening tests confirm that the BLCMV-N leads to a more natural and spatially accurate listening experience for users.

Key Results

  • Objective Performance:
    • Noise Reduction (SNR Improvement): The BLCMV-N achieves a good SNR improvement, though slightly less than the ideal BMVDR. It offers better noise reduction than the BLCMV when the mixing parameter η is increased.
  • Interference Reduction (SIR Improvement): The BLCMV-N provides a high SIR improvement, comparable to the BLCMV, and is primarily controlled by the interference scaling parameter δ.
  • Binaural Cue Preservation (MSC Error): The BLCMV-N significantly reduces the MSC error of the noise component compared to BLCMV, indicating better preservation of background noise spatial cues. This improvement is more pronounced than that of the BMVDR-N.
  • Perceptual Results:
    • Interfering Source: The BLCMV-N, like the BLCMV, significantly preserves the spatial impression of the interfering source, outperforming the BMVDR and BMVDR-N.
  • Background Noise: The BLCMV-N and BMVDR-N significantly preserve the spatial impression of background noise better than the BMVDR and BLCMV.
  • Complete Acoustic Scene: The BLCMV-N achieved the highest scores for preserving the overall spatial impression of the complete acoustic scene, outperforming all other considered algorithms.

Limitations and Future Work

  • Parameter Tuning: While the paper provides optimal settings for maximizing SNR, the optimal choice of η and δ for perceptual quality might require further investigation or user-specific adaptation.
  • Estimation Errors: The performance of the algorithms can be affected by estimation errors in covariance matrices, especially in real-world scenarios.
  • Computational Complexity: While not explicitly detailed, complex beamforming algorithms can have higher computational demands, which is a consideration for real-time implementation in hearing devices.

Future work could explore adaptive parameter selection for η and δ, investigate the performance in more complex acoustic scenarios (e.g., multiple interfering sources), and further optimize the algorithms for computational efficiency.

Summary updated

Abstract

Besides reducing undesired sources, i.e., interfering sources and background noise, another important objective of a binaural beamforming algorithm is to preserve the spatial impression of the acoustic scene, which can be achieved by preserving the binaural cues of all sound sources. While the binaural minimum variance distortionless response (BMVDR) beamformer provides a good noise reduction performance and preserves the binaural cues of the desired source, it does not allow to control the reduction of the interfering sources and distorts the binaural cues of the interfering sources and the background noise. Hence, several extensions have been proposed. First, the binaural linearly constrained minimum variance (BLCMV) beamformer uses additional constraints, enabling to control the reduction of the interfering sources while preserving their binaural cues. Second, the BMVDR with partial noise estimation (BMVDR-N) mixes the output signals of the BMVDR with the noisy reference microphone signals, enabling to control the binaural cues of the background noise. Aiming at merging the advantages of both extensions, in this paper we propose the BLCMV with partial noise estimation (BLCMV-N). We show that the output signals of the BLCMV-N can be interpreted as a mixture between the noisy reference microphone signals and the output signals of a BLCMV using an adjusted interference scaling parameter. We provide a theoretical comparison between the BMVDR, the BLCMV, the BMVDR-N and the proposed BLCMV-N in terms of noise and interference reduction performance and binaural cue preservation. Experimental results using recorded signals as well as the results of a perceptual listening test show that the BLCMV-N is able to preserve the binaural cues of an interfering source (like the BLCMV), while enabling to trade off between noise reduction performance and binaural cue preservation of the background noise (like the BMVDR-N).
BibTeX
@article{goling2020binaural,
  author = {Nico Gößling and Elior Hadad and Sharon Gannot and Simon Doclo},
  title = {{Binaural LCMV Beamforming With Partial Noise Estimation}},
  journal = {IEEE ACM Trans. Audio Speech Lang. Process.},
  year = {2020},
  url = {https://openreview.net/forum?id=ApOzUx3agrf},
}
Figure