French Listening Tests for the Assessment of Intelligibility, Quality, and Identity of Body-Conducted Speech Enhancement

TL;DR

This paper evaluates the EBEN model for body-conducted speech enhancement using listening tests, showing improvements in speech quality and intelligibility but slight degradation in speaker ID, with correlations between objective metrics and perceived quality.

Contribution

It provides a comprehensive assessment of EBEN on body-conduction sensors with diverse recordings and introduces correlations between objective metrics and human perception.

Findings

EBEN improves speech quality and intelligibility.

Slight degradation in speaker identification for female throat recordings.

Correlation found between STOI and perceived quality.

Abstract

This study evaluates the Extreme Bandwidth Extension Network (EBEN) model on body-conduction sensors through listening tests. Using the Vibravox dataset, we assess intelligibility with a French Modified Rhyme Test, speech quality with a MUSHRA (MUltiple Stimuli with Hidden Reference and Anchor) protocol and speaker identity preservation with an A/B identification task. The experiments involved male and female speakers recorded with a forehead accelerometer, rigid in-ear and throat microphones. The results confirm that EBEN enhances both speech quality and intelligibility. It slightly degrades speaker identification performance when applied to female speakers' throat microphone recordings. The findings also demonstrate a correlation between Short-Time Objective Intelligibility (STOI) and perceived quality in body-conducted speech, while speaker verification using ECAPA2-TDNN aligns well with identification performance. No tested metric reliably predicts EBEN's effect on intelligibility.