Deep Learning based singer identification
EntityUAM. Departamento de Tecnología Electrónica y de las Comunicaciones
SubjectsSingerID; Deep Learning; wav2vec; Telecomunicaciones
NoteMaster Universitario en Deep Learning for Audio and Video Signal Processing
Esta obra está bajo una licencia de Creative Commons Reconocimiento-NoComercial-SinObraDerivada 4.0 Internacional.
It is known that speaker identification is a field with a lot of related research carried out but,when it comes to looking for research developed from singingvoiceinstead of speech,only a few studiescan be found. This difference in the amount of work related to both fields is mainly due to the fact that the spoken voice is simpler and contains a much narrower frequency spectrum than the singingvoice. In this way, this Master's Final Project containsa study to identify singers from their recorded songs. For thispurpose, a more sophisticated system has been developed to facethe increased complexity in the data, being able to discriminateamongsingers.As a previous step to identify the singer, and due to the scarcity of databases of singing voice in the state of the art, the present work also includes the development of an automatic way for creating anovel databaseusing Spotify’s API. The database contains information related to the musical genre,the artist and differentmusical characteristics of the 30 seconds excerpt pre-view song provided by Spotify. The files of the songs have been source separated with the network of the Spleeter application to carry out a source separation and thus be able to work with the processedfile that only contains the singingvoice of the original songs.The developed system has used different feature extractors from the current state of the art using both speech analysis techniques and techniques that are used whenmusical instruments are wanted to be identified in recordings. With these obtained features, some current state of the art classifiers have been fed based on shallow neural networks and speaker identification networks.
Google Scholar:Bustos Manzanet, Laura
This item appears in the following Collection(s)
Showing items related by title, author, creator and subject.
Carracedo Cosme, Jaime
Poor sensitivity of fecal gluten immunogenic peptides and serum antibodies to detect duodenal mucosal damage in celiac disease monitoring Laserna-Mendieta, Emilio J.; Casanova, María José; Arias, Ángel; Arias-González, Laura; Majano, Pedro; Mate, Luis Alberto; Gordillo-Vélez, Carlos Humberto; Jiménez, Mirella; Angueira, Teresa; Tébar-Romero, Emilia; Carrillo-Ramos, María Jesús; Tejero-Bustos, María Ángeles; Gisbert, Javier P.; Santander Vaquero, Cecilio; Lucendo, Alfredo J.
Identification of non-coding genetic variants in samples from hypoxemic respiratory disease patients that affect the transcriptional response to hypoxia Roche, Olga; Deguiz, María Laura; Tiana, María; Galiana-Ribote, Clara; Martínez-Alcázar, Daniel; Rey-Serra, Carlos; Ranz-Ribeiro, Beatriz; Casitas, Raquel; Galera, Raúl; Fernández-Navarro, Isabel; Sánchez-Cuéllar, Silvia; Bernard, Virginie; Ancochea Bermúdez, Julio; Wasserman, Wyeth W.; García Rio, Francisco; Jiménez Cuenca, Benilde; del Peso, Luis