Your voice tells a story – Explore one of the world’s most comprehensive databases on so-called „normal“ and pathological voices – The database was created at the former Institute of Phonetics at Saarland University and is currently hosted and managed by Essen University Hospital.
The Saarbrücken Voice Database contains a clinically validated collection of so-called “normal” and pathological voice recordings – The recordings are structured for various research approaches such as AI developments and/or reproducibility.
The collection of voice recordings is suitable for research approaches on so-called “normal” and pathological phonation. In the latter case, functional and organic aspects of vocal fold vibrations are taken into account.
First, each recording session contains recordings of the vowels [i:], [a:], and [u:] in normal, high and low pitch as well as recordings of the vowels with rising and falling pitch. Second, recordings of the sentence “Guten Morgen, wie geht es Ihnen?” (Good morning, how are you?) were also made in the same session.
The database is organized and labeled for ML workflows. Consistent filenames and structured labels facilitate the training of supervised models.
WAV files provide sound pressure signals and electroglottographic signals for phonation analysis and AI pipelines.
Validated annotations by clinical experts to disease patterns such as amyotrophic lateral sclerosis or vocal fold tumors can be found in the database. I addition, control groups of normal speakers can be created for scientific reliability and reproducibility.
Created and collected at the former Institute of Phonetics, Saarland University, & hosted and managed by Essen University Hospital. Free to use for reproducible voice science.
Narrow down the speaker records using gender, phonatory classification (so-called “normal” voices, pathological voices) and diagnosis to focus your search. The complete dataset is available for download here
| Speaker ID | Gender | Recording Type | Pathology |
|---|---|---|---|
| No results match the current filters. | |||
