Vibravox: A Dataset of French Speech Captured with Body-conduction Audio Sensors

TL;DR

Vibravox provides a public, French body‑conduction microphone dataset with five BCM configurations plus an airborne reference, collected from 188 participants under varied, ambisonics‑based noise conditions. The work establishes baselines for three key tasks— bandwidth extension, speech‑to‑phoneme recognition, and speaker verification—using sensor‑specific EBEN bandwidth extension and wav2vec2./ECAPA2‑based pipelines, and assesses performance in quiet and noisy environments. Findings show EBEN reliably improves intelligibility and phoneme transcription across sensors, but can degrade speaker verification, highlighting sensor and task‑specific trade‑offs. The dataset’s rich multi‑sensor, noise‑rich architecture aims to accelerate robust BCM research and practical communication systems in challenging environments.

Abstract

Vibravox is a dataset compliant with the General Data Protection Regulation (GDPR) containing audio recordings using five different body-conduction audio sensors: two in-ear microphones, two bone conduction vibration pickups, and a laryngophone. The dataset also includes audio data from an airborne microphone used as a reference. The Vibravox corpus contains 45 hours per sensor of speech samples and physiological sounds recorded by 188 participants under different acoustic conditions imposed by a high order ambisonics 3D spatializer. Annotations about the recording conditions and linguistic transcriptions are also included in the corpus. We conducted a series of experiments on various speech-related tasks, including speech recognition, speech enhancement, and speaker verification. These experiments were carried out using state-of-the-art models to evaluate and compare their performances on signals captured by the different audio sensors offered by the Vibravox dataset, with the aim of gaining a better grasp of their individual characteristics.

Figures (19)

• Figure 1: Fully equipped participant with the six audio sensors
• Figure 2: Post-processing filtering process
• Figure 3: Coherence functions of body-conducted sensors in quiet conditions
• Figure 4: Spectrograms of signals recorded by the different Vibravox audio sensors in the speech-clean subset and their corresponding EBEN-enhanced version
• Figure 5: Synchronized spectrograms of multiple signals for the speech enhancement task in noisy conditions. (a) & (d): Synthetically mixed samples, speech comes from speech-clean and noise comes from speechless-noisy respective microphones - (b): Corresponding enhanced samples performed by EBEN models trained on the speech-clean subset - (c): Corresponding enhanced samples performed by EBEN models trained on the synthetically mixed speech-clean and speechless-noisy subsets - (e): Clean speech sample recorded by the headset microphone in the speech-clean subset before mixing - (f): Noise sample recorded by the headset microphone in the speechless-noisy subset before mixing