IKA CI Music Preprocessing Listening Experiment Stimuli (2022)

IKA CI Music Preprocessing Listening Experiment Stimuli (2022)

This dataset contains the audio stimuli that have been presented to both cochlear implant (CI) and normal hearing (NH) listeners in the listening experiments in a study named

“A Subjective Evaluation of Different Music Preprocessing Approaches in CI Listeners”

It comprises the excerpts from the IKA CI Pop Music Dataset (IKA-CI-PMD) (10.5281/zenodo.7060282) both as unprocessed references and as processed versions where different music preprocessing strategies have been
applied.

The IKA CI Pop Music Dataset is a dataset of music excerpts that has especially compiled to evaluate different music signal preprocessing strategies for CI listeners. The excerpts taken are from the MedleyDB multitrack dataset (https://medleydb.weebly.com/) curated by Rachel Bittner et. al..

This dataset is split into a 3-piece “training” set (excerpts T01 to T03) that has been used to familiarize the listeners with the experimental setup, and a 12-piece “test” set (E01 to E12) used in actual experiments.

The following music preprocessing strategies are included:

• HPCA+P: Harmonic/percussive sound separation (HPSS) combined with PCA-based spectral complexity reduction [1]
• Cspl and Dspl: DNN-based remix of the harmonic and percussive portions of 4 source stems [2]
• HALCA: Remix based on a probabilistic model for melody extraction using shift invariant kernels in CQT domain [3] (accompaniment attenuated by 12 dB)
• MT remix: Oracle remixes of multitrack stems (other accompaniment attenuated by 12 dB)

All stimuli are normalized to a loudness level of -27 LUFS. The signals are stored in the lossless FLAC format. The files contain stereo signals, where both channels are identical.

The dataset has been compiled at the Ruhr University Bochum Institute of Communication Acoustics in 2022 by Johannes Gauer (johannes.gauer@rub.de) in collaboration with the fellow researchers Anil Nagathil, Benjamin Lentz, and Rainer Martin. Like MedleyDB and the IKA CI Pop Music Dataset, it is licensed under a Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International License.

For further details on the included excerpts refer to the IKA CI Pop Music Dataset (IKA-CI-PMD) (10.5281/zenodo.7060282).

[1] B. Lentz, A. Nagathil, J. Gauer, and R. Martin, “Harmonic/Percussive sound separation and spectral complexity reduction of music signals for cochlear implant listeners,” in Proc IEEE Int Conf Acoust Speech Signal Process ICASSP, Barcelona, Spain, May 2020, pp. 8713–8717.

[2] J. Gauer, A. Nagathil, K. Eckel, D. Belomestny, and R. Martin, “A versatile deep-neural-network-based music preprocessing and remixing scheme for cochlear implant listeners,” J. Acoust. Soc. Am., vol. 151, no. 5, pp. 2975–2986, May 2022.

[3] B. Fuentes, R. Badeau, and G. Richard, “Harmonic Adaptive Latent Component Analysis of Audio and Application to Music Transcription,” IEEE Trans. Audio Speech Lang. Process., vol. 21, no. 9, pp. 1854–1866, Sep. 2013.