To evaluate the performance of audio signal processing algorithms and train emerging data-driven systems, researchers have introduced an open-access acoustic benchmark designated as the Hearing Instrument Dataset in Various Acoustical Scenarios. While audio researchers and engineers can generate extensive batches of simulated audio data using mathematical models, physically recorded acoustic data provide a far more adequate representation of real-life listening scenarios. The newly developed dataset was created to supply the audio community with physical measurements gathered across multiple realistic acoustical environments.
The experimental configuration centered on a dummy head outfitted with hearing instrument components. Behind-the-ear hearing instrument shells were mounted on the head, with two microphones installed on each individual shell to record arriving sound fields. In addition, receiver-in-canal hearing instrument loudspeakers were inserted into the dummy head’s ears, and microphones positioned at the eardrum locations captured sound pressure delivered to the auditory canal.
To establish sound fields and compute impulse responses, the researchers arranged eight external loudspeakers and two external microphones surrounding the dummy head apparatus. The authors computed complete impulse responses for every possible microphone-loudspeaker pair using swept-sine recordings. This acoustic characterization mapped the path between every external speaker, the external reference microphones, the dual behind-the-ear microphones on each shell, and the microphones positioned at the eardrums.
Beyond isolated impulse responses, the investigative team produced extensive multichannel audio recordings using a diverse collection of source signals. The selected audio stimuli included male speech, female speech, speech-shaped noise, and a singing voice, as well as instrumental recordings featuring a stringed instrument, a wind instrument, and a percussion instrument. Each audio excerpt was played back through each individual external loudspeaker while all available microphones across the entire setup recorded simultaneously.
The complete set of audio playback trials was repeated across four different hearing instrument dome configurations. These mechanical setups included open domes, semi-open domes, closed domes, and a no-RIC configuration without the receiver-in-canal component. Testing spanned two separate physical rooms exhibiting four distinct reverberation conditions. In the first room, reverberation was measured under three conditions with T30 times of 0.09 seconds, 0.47 seconds, and 0.73 seconds, while a second room provided a longer reverberation condition with a T30 of 1.48 seconds.
The researchers demonstrated how researchers can use the dataset as a modular hearing instrument in a box through three practical example use cases. These demonstrations examined ipsilateral feedback paths, the direct-to-reverberant sound energy ratio between an external microphone positioned close to the sound source versus the behind-the-ear microphones, and the acoustic leakage traveling from external sound sources directly to the eardrum microphones.
The resulting measurements highlighted several notable acoustic phenomena across the tested hardware and environmental conditions. Acoustic feedback and sound leakage measured for the open and semi-open dome configurations were found to be similar to one another, and both were substantially larger than the feedback and leakage observed with the closed dome. Furthermore, the ratio of direct-to-reverberant sound energy in the behind-the-ear hearing instrument microphones was consistently lower than in the external microphone, and this measured difference grew larger as the room reverberation time increased.
By providing synchronized impulse responses and realistic recordings across diverse acoustic environments, dome types, and reverberation times, the researchers have made the complete dataset publicly available to support future algorithm evaluation and machine learning training in hearing instrument development.
Subject of Research: Acoustic dataset for hearing instrument algorithm evaluation and training
Article Title: HIDVAS: a hearing instrument dataset in various acoustical scenarios for algorithm evaluation and training
Article References: Roebben, A., Bernardi, G., Wouters, J., van Waterschoot, T., & Moonen, M. (2026). HIDVAS: a hearing instrument dataset in various acoustical scenarios for algorithm evaluation and training. Journal on Audio, Speech, and Music Processing. https://doi.org/10.1186/s13636-026-00486-x
Image Credits: AI Generated
DOI: 10.1186/s13636-026-00486-x
Keywords: hearing instruments, acoustic scenarios, algorithm evaluation, impulse responses, dummy head, behind-the-ear microphones, receiver-in-canal loudspeakers, eardrum microphones, swept-sine recordings, room reverberation, acoustic feedback, sound leakage, direct-to-reverberant sound energy, hearing instrument domes, speech-shaped noise, audio signal processing, audio recordings
Cite Scienmag News
Blake Davidson. (November 16, 2026). HIDVAS Dataset Aids Hearing Instrument Algorithm Evaluation and Training. Scienmag. https://scienmag.com/hidvas-dataset-aids-hearing-instrument-algorithm-evaluation-and-training/
Blake Davidson. "HIDVAS Dataset Aids Hearing Instrument Algorithm Evaluation and Training." Scienmag, 16 November 2026, https://scienmag.com/hidvas-dataset-aids-hearing-instrument-algorithm-evaluation-and-training/. Accessed 16 November 2026.
Blake Davidson. "HIDVAS Dataset Aids Hearing Instrument Algorithm Evaluation and Training." Scienmag. November 16, 2026. https://scienmag.com/hidvas-dataset-aids-hearing-instrument-algorithm-evaluation-and-training/








