Audio & Speech

HIDVAS dataset captures 4 hearing aid domes across 4 reverberation levels

New open-source dataset simulates real hearing aid scenarios with 8 speakers, 2 microphones, and a dummy head.

Deep Dive

HIDVAS is introduced as a hearing instrument dataset containing impulse responses and audio recordings from a dummy head equipped with behind-the-ear (BTE) shells and receiver-in-canal (RIC) loudspeakers. Recordings include 7 audio types—male and female speech, speech shaped noise, singing voice, stringed instrument, wind instrument, and percussion instrument—across 4 reverberation conditions (T30 = 0.09 s, 0.47 s, 0.73 s, and 1.48 s) and 4 dome types (open, semi-open, closed, and no-RIC). The dataset supports algorithm evaluation and training, with three example use cases demonstrating its application.

Key Points
  • Dataset includes impulse responses and audio recordings from a dummy head with behind-the-ear and receiver-in-canal hearing instruments.
  • Four hearing aid dome types (open, semi-open, closed, no-RIC) tested across four reverberation conditions (T30 from 0.09s to 1.48s).
  • Includes six audio types (male/female speech, noise, singing, strings, percussion) recorded via eight external loudspeakers and multiple microphones.

Why It Matters

Provides realistic, open-source training data for hearing aid algorithms to improve performance in diverse acoustic environments.

📬 Get the top 10 AI stories daily