Reference Microphone Selection for Guided Source Separation based on the Normalized L-p Norm

2025/10/31 by Lohmann, Anselm, Nakatani, Tomohiro, Ikeshita, Rintaro +3
#Audio and Speech Processing (eess.AS) #FOS: Computer and information sciences #FOS: Electrical engineering #Sound (cs.SD) #electronic engineering #information engineering

paper · doi:10.48550/arxiv.2510.27198

Abstract

Guided Source Separation (GSS) is a popular front-end for distant automatic speech recognition (ASR) systems using spatially distributed microphones. When considering spatially distributed microphones, the choice of reference microphone may have a large influence on the quality of the output signal and the downstream ASR performance. In GSS-based speech enhancement, reference microphone selection is typically performed using the signal-to-noise ratio (SNR), which is optimal for noise reduction but may neglect differences in early-to-late-reverberant ratio (ELR) across microphones. In this paper, we propose two reference microphone selection methods for GSS-based speech enhancement that are based on the normalized ℓ_p-norm, either using only the normalized ℓ_p-norm or combining the normalized ℓ_p-norm and the SNR to account for both differences in SNR and ELR across microphones. Experimental evaluation using a CHiME-8 distant ASR system shows that the proposed ℓ_p-norm-based methods outperform the baseline method, reducing the macro-average word error rate.

Citations

• VERSA: A Versatile Evaluation Toolkit for Speech, Audio, and Music
• SCOREQ: Speech Quality Assessment with Contrastive Regression
• Reference Microphone Selection for the Weighted Prediction Error Algorithm using the Normalized L-p Norm
• STCON System for the CHiME-8 Challenge
• The CHiME-8 DASR Challenge for Generalizable and Array Agnostic Distant Automatic Speech Recognition and Diarization
• TorchAudio-Squim: Reference-less Speech Quality and Intelligibility measures in TorchAudio
• DNSMOS P.835: A Non-Intrusive Perceptual Objective Speech Quality Metric to Evaluate Noise Suppressors
• Learning to Rank Microphones for Distant Speech Recognition
• Comparing Measures of Sparsity

Related

• Guided Source Separation Meets a Strong ASR Backend: Hitachi/Paderborn University Joint Investigation for Dinner Party ASR
• Reference Microphone Selection for the Weighted Prediction Error Algorithm using the Normalized L-p Norm
• Computing Optimal Location of Microphone for Improved Speech Recognition
• Analyzing the impact of speaker localization errors on speech separation for automatic speech recognition
• Transcription-Free Fine-Tuning of Speech Separation Models for Noisy and Reverberant Multi-Speaker Automatic Speech Recognition