Automatic Description of Music in Natural Estonian
Name
Villem Tõnisson
Abstract
Recording and creating music is easier than ever and thanks to the internet, much of this music is publicly available. Due to the difficulty and time consumption of describing songs manually, automatic systems are necessary that could do it for us. The field of music information retrieval (MIR) can help us with this problem. MIR focuses on extracting useful features from music. Most of these features are not usable directly by humans. Natural language descriptors that are created based on these features could provide users with easily understandable ways to receive information about music.
We provide an overview of several tasks in the field of MIR. Our research shows that neural networks are state-of-the-art for almost all looked at tasks in music information retrieval. As a result of this thesis a web application was created that performs tempo estimation, chord recognition, key recognition, form discovery, genre classification and instrument recognition. Based on the retrieved information a description in natural Estonian is generated for the user. We found that the only feasible way to create descriptions was by using templates. For fully automatic description generation, more descriptive textual data about music is necessary.
We provide an overview of several tasks in the field of MIR. Our research shows that neural networks are state-of-the-art for almost all looked at tasks in music information retrieval. As a result of this thesis a web application was created that performs tempo estimation, chord recognition, key recognition, form discovery, genre classification and instrument recognition. Based on the retrieved information a description in natural Estonian is generated for the user. We found that the only feasible way to create descriptions was by using templates. For fully automatic description generation, more descriptive textual data about music is necessary.
Graduation Thesis language
English
Graduation Thesis type
Master - Computer Science
Supervisor(s)
Sven Aller
Defence year
2021