During my master at University of Sheffield I had the crazy idea to do in my dissertation an ASR system in music. Besides all the intrinsic difficult that is an ASR have, I found that, even though several corpus are available for music research such as Medleydb and Million Song Dataset, there are none corpus of music annotated for speech research.

Therefore, I accepted the challenge of creating a new corpus that would satisfy my research needs. As a result of that, Acomus corpus was born.


Please, take into account that this corpus was designed for academic purposes only. Any commercial use or other distinct than academic is strictly forbidden and please avoid to do it.


Acomus is a collection of acoustic covers of popular music that are share on YouTube by amateurs artist. For simplicity, the selected songs have only one accompaniment instrument which in the majority of the set is guitar and in minor proportion is piano.
The collection is base only in adults singers with a reasonable gender balance.

How to get the corpus

Please, if you want to know more about this corpus, please do not hesitate to contact me, or follow this link ACOMUS.