Published July 13, 2025 | Version v1
Conference proceeding Open

MB-RIRs: a Synthetic Room Impulse Response Dataset with Frequency-Dependent Absorption Coefficients

  • 1. Universitat Pompeu Fabra, Music Technology Group
  • 2. Eurecat, Centre Tecnològic de Catalunya, Tecnologies Multimèdia

Description

We investigate the effects of four strategies for improving the ecological validity of synthetic room impulse response (RIR) datasets for monoaural Speech Enhancement (SE). We implement three features on top of the traditional image source method-based (ISM) shoebox RIRs: multiband absorption coefficients, source directivity and receiver directivity. We additionally consider mesh-based RIRs from the SoundSpaces dataset. We then train a DeepFilternet3 model for each RIR dataset and evaluate the performance on a test set of real RIRs both objectively and subjectively. We find that RIRs which use frequency-dependent acoustic absorption coefficients (MB-RIRs) can obtain +0.51dB of SDR and a +8.9 MUSHRA score when evaluated on real RIRs. The MB-RIRs dataset is publicly available for free download.

Files

2507.09750v1.pdf

Files (416.2 kB)

Name Size Download all
md5:c1075d993b38137b8a46d4255f9ac12e
416.2 kB Preview Download

Additional details

Funding

European Commission
GuestXR - GuestXR: A Machine Learning Agent for Social Harmony in eXtended Reality 101017884