[Paper Review] Optimal Binaural LCMV Beamforming in Complex Acoustic Scenarios: Theoretical and Practical Insights

Nico Gößling, Daniel Marquardt+3 more|arXiv (Cornell University)|Jul 12, 2018

Speech and Audio Processing14 references4 citations

TL;DR

This paper proposes optimal interference scaling parameters for the binaural linearly constrained minimum variance (BLCMV) beamformer to preserve binaural cues in complex acoustic scenarios. By applying upper and lower thresholds to these parameters, the method achieves robust performance despite estimation errors from short observation intervals, significantly reducing binaural cue errors while maintaining effective noise reduction in real-world hearing aid applications using measured impulse responses.

ABSTRACT

Binaural beamforming algorithms for head-mounted assistive listening devices are crucial to improve speech quality and speech intelligibility in noisy environments, while maintaining the spatial impression of the acoustic scene. While the well-known BMVDR beamformer is able to preserve the binaural cues of one desired source, the BLCMV beamformer uses additional constraints to also preserve the binaural cues of interfering sources. In this paper, we provide theoretical and practical insights on how to optimally set the interference scaling parameters in the BLCMV beamformer for an arbitrary number of interfering sources. In addition, since in practice only a limited temporal observation interval is available to estimate all required beamformer quantities, we provide an experimental evaluation in a complex acoustic scenario using measured impulse responses from hearing aids in a cafeteria for different observation intervals. The results show that even rather short observation intervals are sufficient to achieve a decent noise reduction performance and that a proposed threshold on the optimal interference scaling parameters leads to smaller binaural cue errors in practice.

Motivation & Objective

• To derive optimal interference scaling parameters for the BLCMV beamformer that preserve binaural cues of both desired and interfering sources.
• To address performance degradation in practice due to estimation errors from short temporal observation intervals.
• To propose a thresholding strategy on interference scaling parameters to enhance robustness against RTF estimation errors.
• To evaluate the performance of BMVDR and BLCMV beamformers using different correlation matrices and observation intervals in a real-world cafeteria scenario.
• To demonstrate that short observation intervals are sufficient for effective noise reduction and binaural cue preservation when using thresholded parameters.

Proposed method

• Derives optimal interference scaling parameters for BLCMV based on the BMVDR-RTF beamformer, ensuring binaural cue preservation for multiple interfering sources.
• Applies upper and lower thresholds to the optimal scaling parameters to improve robustness against RTF estimation errors in practice.
• Uses measured impulse responses from hearing aids in a cafeteria to simulate realistic complex acoustic scenarios.
• Employs a weighted overlap-add processing framework with 256-point blocks and 50% overlap for real-time beamformer implementation.
• Estimates correlation matrices (R_y, R_v, R_n) and RTFs from short temporal observation intervals to assess performance under estimation uncertainty.
• Evaluates beamformers using SINR improvement and binaural cue errors (ILD and ITD) as performance metrics, averaged across frequencies and scenarios.

Experimental results

Research questions

• RQ1 What are the optimal interference scaling parameters for the BLCMV beamformer that preserve binaural cues of multiple interfering sources?
• RQ2 How do estimation errors from short observation intervals affect the performance of BLCMV and BMVDR beamformers?
• RQ3 Can thresholding the optimal interference scaling parameters improve robustness against RTF estimation errors in practice?
• RQ4 How does the choice of correlation matrix (R_y, R_v, R_n) impact SINR improvement and binaural cue preservation?
• RQ5 Is there a trade-off between SINR improvement and binaural cue error when using optimal vs. thresholded interference scaling parameters?

Key findings

• Even short observation intervals (below 200 ms) yield sufficient noise reduction performance, indicating that estimation from limited data is viable.
• Using the optimal interference scaling parameters (δ^opt) in BLCMV leads to binaural cue errors similar to those of the BMVDR beamformer, especially at small observation intervals.
• Applying thresholds to the optimal interference scaling parameters (δ^thr) significantly reduces binaural cue errors, particularly for ITD, compared to δ^opt.
• The BLCMV beamformer with δ^thr outperforms BMVDR in SINR for observation intervals above 300 ms when using R_n, due to additional constraints.
• Using R_n instead of R_v is recommended in dynamic scenarios with short observation intervals, as R_v is difficult to estimate accurately.
• Informal listening tests confirm that δ^thr leads to better perceptual quality by preserving spatial cues more effectively.