Audio Systems Lab Experimental Data
Kevin D. Donohue

Center for Visualization and Virtual Environments
Audio Systems Laboratory
University
of Kentucky
(Last Update 08-15-2009)

 

This page is a directory to experimental data that has potential for reuse in other projects. It is primarily for students doing research with signal processing and audio array algorithms. All are welcome to used the data. If used in a publication/document, it should be acknowledged. The data cannot be used in a commercial product. Documentation is provided with the data. A significant amount of Matlab functions and scripts have been written to work with this data. Many of these can be found in the ArrayToolbox. If there are questions or comments, please send them along (donohue@engr.uky.edu). I would be interested in hearing about your applications.

 

Data Sets:

Cocktail Party Recordings

Single-Track Single-Speaker Speech

 

Cocktail party recording with distributed microphones (Speaker of interest embedded in many simultaneous speakers)

CocktailParty_Data - This document describes the data for the cocktail party simulations. The data were recorded so the cocktail party noise and the speaker of interest (SOI) can be separated and combined at various signal to noise (SNR) levels. This enables the development and more efficient testing of beamforming algorithms, especially for microphone arrays with irregular spatial distributions recording in a cocktail party environment.

bftest - This Matlab script provides an example for reading in experimental data files associated with a series of cocktail recordings. Experimental scenarios were recorded with spatially distributed microphones where the speaker of interest (SOI) was recorded separately from the party of multiple simultaneous conversations. The microphone geometry and settings were the same for both recordings, so they can be superimposed at specified power ratios and/or processed separately for performance metric estimations, such as SNR and intelligibility. Additionally, this script requires several of the scripts that are found in the ArrayToolbox.

The following cocktail party simulations were performed. Each image shows the locations of the microphones in the cube with the blue circles. The cluster of red X's shows the positions of one of the SOIs in the data set.. Click on the image to download the data files using that microphone arrangement. The data files can then be used in conjunction with the Matlab script above and the Array Toolbox to see the results using that specific arrangement.

 

spreadcluster
05/29/08 Spread Cluster

wallplane
06/04/08 Wall Plane

linear
07/01/08 Linear

3dgrid
07/22/08 3D Grid

cornercluster
08/01/08 Corner Cluster

pairs3D
08/22/08 Pairwise Even 3D

audioexpdata_htm_190dd939
10/15/08 Ceiling Plane


02/13/09 Ceiling Regular Rectangular Grid

hex_ceil.jpg
02/23/09 Ceiling Regular Hexagonal Grid

ceilrand031709.jpg

03/17/09 Ceiling Irregular Array

ceilrand040909.jpg
04/04/09 Ceiling Irregular Array

linear041509.jpg
04/04/09 Ceiling Irregular Array

perimeterreg.jpg
04/24/09 Regular Perimeter Array


05/14/09 Endfire Clustered Array

 

 

 

 

 

 

Single-Track Single-Speaker Speech

 

The following links are zipped directories with single-person single-microphone recordings in a relatively quiet office environment. The recording was made with a Behringer ECM 8000 omnidirectional measurement microphone connected to an M-Audio Audio Buddy preamp, and digitized through a single channel of an M-Audio Delta 1010 Digitizer at 44.1kHz sampling at 24bits per sample. The speaker was approximately 9 to 18 inches from the microphone and shielded on two sides (behind the microphone) by acoustic treatments to limit reverberations. Low level noise and room modes were reduced through post-filtering performed using a spectral envelope noise reduction algorithm in Goldwave. The file was finally saved as a 16-bit mono wave file.

 

Each folder contains speech from a single individual reading a script dominated by words used in intelligibility studies with children. There may be up to 3 wave files in each folder (about 20 seconds long) with corresponding text files containing a transcript of actual words/sounds made by the speaker. These files are useful for signal processing, word recognition, and intelligibility experiments. They can be used directly in the cocktail party simulators in the Audio Array Toolbox that simulate positions for the speaker(s) and microphone as well as include noise and room effects.

 

Adult Male Voices

Adult Female Voices

man1

woman1

man2

woman2

man3

woman3

man4

woman4

man5

man6

man7

man8