A Corpus-Based Evaluation of Beamforming Techniques and Phase-Based Frequency Masking

Area of Science:

• Signal Processing
• Acoustics
• Machine Learning

Background:

• Beamforming is crucial for audio array processing, enabling interference reduction, sound source localization, and pre-processing for classification tasks.
• The auditory scene analysis community requires systematic comparisons of various beamforming techniques to guide method selection.

Purpose of the Study:

• To conduct a comprehensive evaluation and comparison of five popular beamforming techniques.
• To assess the performance of a novel phase-based frequency masking beamformer against established methods.
• To provide a reproducible framework and open-access resources for the research community.

Main Methods:

• Evaluation of five beamforming techniques using the Acoustic Interactions for Robot Audition (AIRA) corpus.
• Systematic variation of experimental parameters including microphone count, interference levels, and direction-of-arrival error.
• Implementation and assessment of a phase-based frequency masking beamformer within a common software framework.

Main Results:

• The phase-based frequency masking beamformer demonstrated superior performance compared to the five evaluated techniques across various acoustic conditions.
• Performance variations were observed based on the number of microphones, interference levels, and direction-of-arrival accuracy.
• The study provides detailed insights into the tendencies and effectiveness of different beamforming approaches.

Conclusions:

• The phase-based frequency masking beamformer represents a significant advancement in audio array processing.
• The open availability of the corpus and implementations promotes transparency and repeatability in auditory scene analysis research.
• Informed decisions regarding beamforming technique selection can be made based on the presented evaluation results and observed tendencies.