BUT Speech@FIT Reverb Database

This is the first release of BUT Speech@FIT Reverb Database. The database is being built with respect to collect a large number of various Room Impulse Responses, Room environmental noises (or "silences"), Retransmitted speech (for ASR and SID testing), and meta-data (positions of microphones, speakers etc.).

The goal is to provide speech community with a dataset for data enhancement and distant microphone or microphone array experiments in ASR and SID.

The database has CC-BY 4.0 license and you can download it here:

The BUT Speech@FIT Reverb Dataset consists of 9 rooms:

Size [m x m x m] Volume [m^3] # RIRs Ret. Type In RIR-Only set In LibriSpeech-Only set
Q301 10.7x6.9x2.6 192 31 x 3 1 Office Yes Yes
L207 4.6x6.9x3.1 98 31 x 6 3 Office Yes Yes
L212 7.5x4.6x3.1 107 31 x 5 2 Office Yes Yes
L227 6.2x2.6x14.2 229 31 x 5 3 Stairs Yes Yes
R112 4.4x2.8x2.6* ~40 31 x 5 0 Hotel room Yes No
CR2 28.2x11.1x3.3 1033 31 x 4 0 Conf. room Yes No
E112 11.5x20.1x4.8* ~900 31 x 2 0 Lect. room Yes No
D105 17.2x22.8x6.9* ~2000 31 x 6 1 Lect. room Yes Yes
C236 7.0x4.1x3.6 102 31 x 10 0 Meeting room Yes No

We placed 31 microphones in all rooms. The source (a Hi-Fi loudspeaker) was placed on 5 positions in average. We measured RIRs (using exponential sine sweep method) for each speaker position. Next we recorded environmental noise (silence). There was a radio at background playing in one speaker position in the office.

We also retransmitted LibriSpeech Test-clean dataset for some of the positions of speaker (column Ret. in the table above). This data is freely available from our web-pages along with the RIRs. We also retransmitted a portion of NIST Speaker recognition evaluation 2010 dataset, and HUB5 2000 RT eval set. The availability of this data is limited to sites that have valid LDC license to the original data.

All microphone positions are measured and stored in meta-files. We pre-calculated positions of microphones and speakers in Cartesian and polar coordinates as absolute and relative (to the speaker).

Please see attached README.txt for more detailed description of data.

If you want to publish a paper using this dataset, please cite: https://ieeexplore.ieee.org/document/8717722 (DOI:10.1109/JSTSP.2019.2917582, https://arxiv.org/abs/1811.06795) and refer to this page. Recipe for experiments reported in the paper is here: AMI_Kaldi_recipe.tar.gz

Feel free to provide us with your feedback to szoke@fit.vutbr.cz with a subject mentioning BUT-ReverbDB.