CSL-EMG_Array Corpus

The CSL-EMG_Array corpus is an open access parallel EMG-Audio corpus meant for the investigation of EMG-to-Speech conversion. It is available under the Creative Commons Attribution 4.0 license.

Data overview

The CSL-EMG_Array corpus contains parallel facial surface EMG and Audio data during audible as well as silent speech production, along with metadata (most importantly, recording times and text). The EMG data was recorded with a 4x8 + 1x8 array sEMG setup, while the audio data was recorded with a high quality condenser microphone. Both signals contain a marker channel that can be used for synchronization. There are 12 sessions by 8 speakers available: 8 audible sessions (all data spoken audibly) and 4 silent sessions (training data spoken audibly, evaluation data spoken silently). In total, the corpus contains approximately 9.7 hours of data. Signals are provided as numpy version 3 data files, while the metadata is provided in a text based json format. The total download size is 8.8 Gigabytes. Detailed information on the corpus composition and data formats as well as baseline results can be found in the accompanying paper: CSL-EMG_Array: An Open Access Corpus for EMG-to-Speech Conversion

Corpus download

The corpus can be downloaded here: CSL-EMG_Array corpus download. After entering your e-mail, a download link will be sent to you.

How to cite

When you use this corpus in your work, you agreed to the Creative Commons Attribution 4.0 International License, which means that you have to properly cite the CSL-EMG_Array Corpus - please use the following publication:

Lorenz Diener, Mehrdad Roustay Vishkasougheh, Tanja Schultz (Cognitive Systems Lab, University of Bremen, Germany), "CSL-EMG_Array: An Open Access Corpus for EMG-to-Speech Conversion", INTERSPEECH 2020

You may wish to use the following bibtex entry:

@inproceedings{diener2020cslemgarray,
    title={{CSL-EMG\_Array: An Open Access Corpus for EMG-to-Speech Conversion}},
    author={Diener, Lorenz and  Roustay Vishkasougheh, Mehrdad and Schultz, Tanja},
    booktitle={{INTERSPEECH} 2020 -- 21st Annual Conference of the International Speech Communication Association},
    year={2020},
}

Acknowledgements

The authors would like to thank Google for the generous gift used towards “2020 Faculty Research Award in recognition of Dr. Tanja Schultz’s research related to EMG-to-Speech: End-to-End Silent Speech Conversion based on Electromyography”, which greatly helped us in performing this data collection. We hope that our published results and open source data corpus can form the basis for much further research into EMG-to-Speech conversion.