Speech Enhancement and Assessment Resource (SpEAR)

The SpEAR Datbase contains carefully selected samples of noise corrupted speech with clean speech references. This beta version is a small initial release of the database to allow feedback for future developments of this resource.

All speech and noise sources have been acoustically combined and re-recorded. Synchronous clocking is used to provide an exact time-aligned reference to the clean speech signal (see Technical Details).

Contents (click on link for details and table of wavfiles):

• Noisy Speech Recordings - Recorded speech and recorded noise: acoustically combined and re-recorded. Various noise sources and different SNR levels.

• Lombard Speech - Live speech in a noisy environment. Recorded noise sources are used to allow extraction of a cleaned speech reference.

• Monaural Recordings - Two recorded speech signals: acoustically combined and re-recorded.

Future releases will contain a more substantial number of samples, additional additive noise sources, cellular and coding noise, etc.

Use of this data is free, but should be referenced in all publications as follows:

Speech Enhancement Assessment Resource (SpEAR) Database. http://ee.ogi.edu/NSEL/. Beta Release v1.0. CSLU, Oregon Graduate Institute of Science and Technology. E. Wan, A. Nelson, and Rick Peterson.