A machine learning approach to violin timbre quality classification

TitleA machine learning approach to violin timbre quality classification
Publication TypeMaster Thesis
Year of Publication2018
AuthorsLladó, P.
AbstractTimbre definition has been traditionally unclear and it is difficult to find a precise definition related to timbre attributes, since several factors contribute to its perception. In this thesis, we present a machine learning approach to automatically classify expert defined timbral attributes, by extracting audio features from violin recordings. Features were extracted using the ESSENTIA library and machine learning models were obtained to classify the aforementioned timbre attributes. Automatic feature selection tools were used to study the most relevant features for classification. Results might indicate that the extracted audio features contain sufficient information to correctly classify the studied timbral attributes.
Keywordsfeature selection, Machine learning, mir, Timbre, Violin
Final publicationhttps://doi.org/10.5281/zenodo.1468983
intranet