Lombard Speech Database for German Language
This database provides a comprehensive collection of German language speech recordings demonstrating the Lombard effect, which refers to involuntary speech adaptation mechanisms human speakers employ to maintain intelligibility in noisy environments. The dataset was created at TU Ilmenau in collaboration with HEAD acoustics GmbH and AVM GmbH, and presented at DAGA 2016 in Aachen.
Description
This database provides a comprehensive collection of German language speech recordings demonstrating the Lombard effect, which refers to involuntary speech adaptation mechanisms human speakers employ to maintain intelligibility in noisy environments. The dataset was created at TU Ilmenau in collaboration with HEAD acoustics GmbH and AVM GmbH, and presented at DAGA 2016 in Aachen.
The database features recordings from 8 German native speakers (5 male, 3 female) who were presented with three different noise conditions via headphones, eliciting varying degrees of raised voice volume and specifically stressed speech characteristics typical of Lombard speech. The noise stimulus used during recordings was artificially created babble speech consisting of multiple overlaid speech recordings, simulating realistic conversational noise environments.
An important technical characteristic of this database is that due to headphone presentation of the noise stimuli, the speech recordings themselves do not include the noise signals. This design allows researchers to mix the clean Lombard speech with different types of noises during post-processing, providing flexibility for various speech intelligibility and enhancement studies. This approach enables controlled experimentation with different signal-to-noise ratios and noise types while maintaining the authentic Lombard speech characteristics.
The Lombard effect research is significant because speech produced in noise (Lombard speech) demonstrates measurably higher intelligibility than speech produced in quiet conditions (plain speech). Systems designed to operate in noisy environments benefit substantially from training on well-matched Lombard speech data, showing significant improvements in automatic speech recognition and communication systems. The database has been utilized in research on speech intelligibility enhancement, voice conversion, and the development of noise-adaptive speech processing algorithms.
This resource is particularly valuable for researchers working on speech intelligibility in challenging acoustic conditions, development of speech enhancement algorithms, training of automatic speech recognition systems for noisy environments, and studies on human speech adaptation mechanisms. The database is openly available on Zenodo under Creative Commons Attribution 4.0 International License, supporting reproducible research in speech science and audio engineering.
Access
Openly available for download from Zenodo (761.7 MB main file, 176.0 GB total)
License
Creative Commons Attribution 4.0 International