Implementation of active data Selection algorithms for data choosing in ASV systems
In the field of voice spoof detection, training countermea- sures (CM) that effectively generalize to various unseen tests is a per- sistent challenge. While strategies such as data augmentation and self- supervised learning have been employed to enhance CM, their limited performance still requires...
- Autores:
-
Jiménez Garizao, Jesús Alberto
- Tipo de recurso:
- Trabajo de grado de pregrado
- Fecha de publicación:
- 2023
- Institución:
- Universidad de los Andes
- Repositorio:
- Séneca: repositorio Uniandes
- Idioma:
- eng
- OAI Identifier:
- oai:repositorio.uniandes.edu.co:1992/73644
- Acceso en línea:
- https://hdl.handle.net/1992/73644
- Palabra clave:
- Active learning
Voice anti-spoofing
Countermeasure
Logical access
Deep learning
Ingeniería
- Rights
- embargoedAccess
- License
- Attribution 4.0 International
| id |
UNIANDES2_a43bcc15ffb47504df4944c362a2d41a |
|---|---|
| oai_identifier_str |
oai:repositorio.uniandes.edu.co:1992/73644 |
| network_acronym_str |
UNIANDES2 |
| network_name_str |
Séneca: repositorio Uniandes |
| repository_id_str |
|
| dc.title.eng.fl_str_mv |
Implementation of active data Selection algorithms for data choosing in ASV systems |
| title |
Implementation of active data Selection algorithms for data choosing in ASV systems |
| spellingShingle |
Implementation of active data Selection algorithms for data choosing in ASV systems Active learning Voice anti-spoofing Countermeasure Logical access Deep learning Ingeniería |
| title_short |
Implementation of active data Selection algorithms for data choosing in ASV systems |
| title_full |
Implementation of active data Selection algorithms for data choosing in ASV systems |
| title_fullStr |
Implementation of active data Selection algorithms for data choosing in ASV systems |
| title_full_unstemmed |
Implementation of active data Selection algorithms for data choosing in ASV systems |
| title_sort |
Implementation of active data Selection algorithms for data choosing in ASV systems |
| dc.creator.fl_str_mv |
Jiménez Garizao, Jesús Alberto |
| dc.contributor.advisor.none.fl_str_mv |
Manrique Piramanrique, Rubén Francisco |
| dc.contributor.author.none.fl_str_mv |
Jiménez Garizao, Jesús Alberto |
| dc.contributor.researchgroup.none.fl_str_mv |
Facultad de Ingeniería |
| dc.subject.keyword.eng.fl_str_mv |
Active learning Voice anti-spoofing Countermeasure Logical access Deep learning |
| topic |
Active learning Voice anti-spoofing Countermeasure Logical access Deep learning Ingeniería |
| dc.subject.themes.spa.fl_str_mv |
Ingeniería |
| description |
In the field of voice spoof detection, training countermea- sures (CM) that effectively generalize to various unseen tests is a per- sistent challenge. While strategies such as data augmentation and self- supervised learning have been employed to enhance CM, their limited performance still requires additional approaches. This research focuses on the use of active learning (AL) to select and eliminate training data in CM training, addressing the need to optimize data selection and im- prove model effectiveness. The study proposes several active learning algorithms that offer substantial improvements in detection error rates across multiple datasets in automatic speaker verification (ASV) sys- tems, these are based on the ASVspoof 2019 and the HABLA sets. Thus, these contributions are expected to be valuable for future research and applications in this domain, significantly enhancing model effectiveness. |
| publishDate |
2023 |
| dc.date.issued.none.fl_str_mv |
2023-12-04 |
| dc.date.accessioned.none.fl_str_mv |
2024-01-30T21:54:42Z |
| dc.date.accepted.none.fl_str_mv |
2024-01-27 |
| dc.date.available.none.fl_str_mv |
2025-01-25 |
| dc.type.none.fl_str_mv |
Trabajo de grado - Pregrado |
| dc.type.driver.none.fl_str_mv |
info:eu-repo/semantics/bachelorThesis |
| dc.type.version.none.fl_str_mv |
info:eu-repo/semantics/acceptedVersion |
| dc.type.coar.none.fl_str_mv |
http://purl.org/coar/resource_type/c_7a1f |
| dc.type.content.none.fl_str_mv |
Text |
| dc.type.redcol.none.fl_str_mv |
http://purl.org/redcol/resource_type/TP |
| format |
http://purl.org/coar/resource_type/c_7a1f |
| status_str |
acceptedVersion |
| dc.identifier.uri.none.fl_str_mv |
https://hdl.handle.net/1992/73644 |
| dc.identifier.instname.none.fl_str_mv |
instname:Universidad de los Andes |
| dc.identifier.reponame.none.fl_str_mv |
reponame:Repositorio Institucional Séneca |
| dc.identifier.repourl.none.fl_str_mv |
repourl:https://repositorio.uniandes.edu.co/ |
| url |
https://hdl.handle.net/1992/73644 |
| identifier_str_mv |
instname:Universidad de los Andes reponame:Repositorio Institucional Séneca repourl:https://repositorio.uniandes.edu.co/ |
| dc.language.iso.none.fl_str_mv |
eng |
| language |
eng |
| dc.relation.references.none.fl_str_mv |
Xin Wang and Junichi Yamagishi. A Practical Guide to Logical Access Voice Presentation Attack Detection, pages 169–214. Springer Nature Singapore, Singapore, 2022. Rohan Das, Jichen Yang, and Haizhou Li. Assessing the scope of generalized countermeasures for anti-spoofing. 05 2020. Burr Settles. Active learning literature survey. 07 2010. Rohan Kumar Das, Jichen Yang, and Haizhou Li. Data augmentation with signal companding for detection of logical access attacks. In ICASSP 2021 - 2021 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP), pages 6349–6353, 2021. Xin Wang and Junichi Yamagishi. Investigating active-learning-based training data selection for speech spoofing countermeasure. In 2022 IEEE Spoken Language Technology Workshop (SLT), pages 585–592, 2023. Gavin C. Cawley. Baseline methods for active learning. In Isabelle Guyon, Gavin Cawley, Gideon Dror, Vincent Lemaire, and Alexander Statnikov, editors, Active Learning and Experimental Design workshop In conjunction with AISTATS 2010, volume 16 of Proceedings of Machine Learning Research, pages 47–57, Sardinia, Italy, 16 May 2011. PMLR. Josh Attenberg and Foster Provost. Inactive learning? difficulties employing active learning in practice. SIGKDD Explorations, 12:36–41, 01 2010. Pablo Andrés Tamayo Flórez, Rubén Manrique, and Bernardo Pereira Nunes. HABLA: A Dataset of Latin American Spanish Accents for Voice Anti-spoofing. In Proc. INTERSPEECH 2023, pages 1963–1967, 2023. Akshay L. Chandra, Sai Vikas Desai, Chaitanya Devaguptapu, and Vineeth N. Balasubramanian. On initial pools for deep active learning. In Luca Bertinetto, Joaõ F. Henriques, Samuel Albanie, Michela Paganini, and Gül Varol, editors, NeurIPS 2020 Workshop on Pre-registration in Machine Learning, volume 148 of Proceedings of Machine Learning Research, pages 14–32. PMLR, 11 Dec 2021. Xin Wang and Junichi Yamagishi. Investigating Self-Supervised Front Ends for Speech Spoofing Countermeasures. In Proc. The Speaker and Language Recognition Workshop (Odyssey 2022), pages 100–106, 2022. Massimiliano Todisco, Xin Wang, Ville Vestman, Md Sahidullah, Hector Del- gado, Andreas Nautsch, Junichi Yamagishi, Nicholas Evans, Tomi Kinnunen, and Kong Aik Lee. Asvspoof 2019: Future horizons in spoofed and fake audio detection, 2019. |
| dc.rights.en.fl_str_mv |
Attribution 4.0 International |
| dc.rights.uri.none.fl_str_mv |
http://creativecommons.org/licenses/by/4.0/ |
| dc.rights.accessrights.none.fl_str_mv |
info:eu-repo/semantics/embargoedAccess |
| dc.rights.coar.none.fl_str_mv |
http://purl.org/coar/access_right/c_f1cf |
| rights_invalid_str_mv |
Attribution 4.0 International http://creativecommons.org/licenses/by/4.0/ http://purl.org/coar/access_right/c_f1cf |
| eu_rights_str_mv |
embargoedAccess |
| dc.format.extent.none.fl_str_mv |
14 páginas |
| dc.format.mimetype.none.fl_str_mv |
application/pdf |
| dc.publisher.none.fl_str_mv |
Universidad de los Andes |
| dc.publisher.program.none.fl_str_mv |
Ingeniería de Sistemas y Computación |
| dc.publisher.faculty.none.fl_str_mv |
Facultad de Ingeniería |
| dc.publisher.department.none.fl_str_mv |
Departamento de Ingeniería Sistemas y Computación |
| publisher.none.fl_str_mv |
Universidad de los Andes |
| institution |
Universidad de los Andes |
| bitstream.url.fl_str_mv |
https://repositorio.uniandes.edu.co/bitstreams/f93beba8-a0cc-497c-8179-c6bbe3ecc332/download https://repositorio.uniandes.edu.co/bitstreams/f105fe38-de1b-41d6-be6e-32bab4d90fad/download https://repositorio.uniandes.edu.co/bitstreams/1ce538e8-f5d6-4c74-8c80-f78d5a3efab2/download https://repositorio.uniandes.edu.co/bitstreams/7bd0ff29-36ee-4bde-b126-36bc915d696f/download https://repositorio.uniandes.edu.co/bitstreams/0cfaa3e5-6ba0-4ce8-ac70-fc7c80231c85/download https://repositorio.uniandes.edu.co/bitstreams/2990c984-8a58-406d-9939-b5c2180b285d/download https://repositorio.uniandes.edu.co/bitstreams/e8a097ab-8c29-4580-b939-a49b9cf94a74/download https://repositorio.uniandes.edu.co/bitstreams/8871c1b0-4495-42b0-bcdc-1110e2124cd7/download |
| bitstream.checksum.fl_str_mv |
d9921c9e9d7061b558245ddf9bec4ad9 e55ff3fd2c68dc12071bdce45695ee59 0175ea4a2d4caec4bbcc37e300941108 ae9e573a68e7f92501b6913cc846c39f f4ef904ad2c1be221cf2f2ca5424bac7 530d320eb9a3b695b3582e4fc981c68e 734d141c1598fab4b0dce15627b25d96 5811fdeb69405a32c18155d67596c071 |
| bitstream.checksumAlgorithm.fl_str_mv |
MD5 MD5 MD5 MD5 MD5 MD5 MD5 MD5 |
| repository.name.fl_str_mv |
Repositorio institucional Séneca |
| repository.mail.fl_str_mv |
adminrepositorio@uniandes.edu.co |
| _version_ |
1808390279232749568 |
| spelling |
Manrique Piramanrique, Rubén Franciscovirtual::229-1Jiménez Garizao, Jesús AlbertoFacultad de Ingeniería2024-01-30T21:54:42Z2025-01-252023-12-042024-01-27https://hdl.handle.net/1992/73644instname:Universidad de los Andesreponame:Repositorio Institucional Sénecarepourl:https://repositorio.uniandes.edu.co/In the field of voice spoof detection, training countermea- sures (CM) that effectively generalize to various unseen tests is a per- sistent challenge. While strategies such as data augmentation and self- supervised learning have been employed to enhance CM, their limited performance still requires additional approaches. This research focuses on the use of active learning (AL) to select and eliminate training data in CM training, addressing the need to optimize data selection and im- prove model effectiveness. The study proposes several active learning algorithms that offer substantial improvements in detection error rates across multiple datasets in automatic speaker verification (ASV) sys- tems, these are based on the ASVspoof 2019 and the HABLA sets. Thus, these contributions are expected to be valuable for future research and applications in this domain, significantly enhancing model effectiveness.Ingeniero de Sistemas y ComputaciónPregrado14 páginasapplication/pdfengUniversidad de los AndesIngeniería de Sistemas y ComputaciónFacultad de IngenieríaDepartamento de Ingeniería Sistemas y ComputaciónAttribution 4.0 Internationalhttp://creativecommons.org/licenses/by/4.0/info:eu-repo/semantics/embargoedAccesshttp://purl.org/coar/access_right/c_f1cfImplementation of active data Selection algorithms for data choosing in ASV systemsTrabajo de grado - Pregradoinfo:eu-repo/semantics/bachelorThesisinfo:eu-repo/semantics/acceptedVersionhttp://purl.org/coar/resource_type/c_7a1fTexthttp://purl.org/redcol/resource_type/TPActive learningVoice anti-spoofingCountermeasureLogical accessDeep learningIngenieríaXin Wang and Junichi Yamagishi. A Practical Guide to Logical Access Voice Presentation Attack Detection, pages 169–214. Springer Nature Singapore, Singapore, 2022.Rohan Das, Jichen Yang, and Haizhou Li. Assessing the scope of generalized countermeasures for anti-spoofing. 05 2020.Burr Settles. Active learning literature survey. 07 2010.Rohan Kumar Das, Jichen Yang, and Haizhou Li. Data augmentation with signal companding for detection of logical access attacks. In ICASSP 2021 - 2021 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP), pages 6349–6353, 2021.Xin Wang and Junichi Yamagishi. Investigating active-learning-based training data selection for speech spoofing countermeasure. In 2022 IEEE Spoken Language Technology Workshop (SLT), pages 585–592, 2023.Gavin C. Cawley. Baseline methods for active learning. In Isabelle Guyon, Gavin Cawley, Gideon Dror, Vincent Lemaire, and Alexander Statnikov, editors, Active Learning and Experimental Design workshop In conjunction with AISTATS 2010, volume 16 of Proceedings of Machine Learning Research, pages 47–57, Sardinia, Italy, 16 May 2011. PMLR.Josh Attenberg and Foster Provost. Inactive learning? difficulties employing active learning in practice. SIGKDD Explorations, 12:36–41, 01 2010.Pablo Andrés Tamayo Flórez, Rubén Manrique, and Bernardo Pereira Nunes. HABLA: A Dataset of Latin American Spanish Accents for Voice Anti-spoofing. In Proc. INTERSPEECH 2023, pages 1963–1967, 2023.Akshay L. Chandra, Sai Vikas Desai, Chaitanya Devaguptapu, and Vineeth N. Balasubramanian. On initial pools for deep active learning. In Luca Bertinetto, Joaõ F. Henriques, Samuel Albanie, Michela Paganini, and Gül Varol, editors, NeurIPS 2020 Workshop on Pre-registration in Machine Learning, volume 148 of Proceedings of Machine Learning Research, pages 14–32. PMLR, 11 Dec 2021.Xin Wang and Junichi Yamagishi. Investigating Self-Supervised Front Ends for Speech Spoofing Countermeasures. In Proc. The Speaker and Language Recognition Workshop (Odyssey 2022), pages 100–106, 2022.Massimiliano Todisco, Xin Wang, Ville Vestman, Md Sahidullah, Hector Del- gado, Andreas Nautsch, Junichi Yamagishi, Nicholas Evans, Tomi Kinnunen, and Kong Aik Lee. Asvspoof 2019: Future horizons in spoofed and fake audio detection, 2019.202020431Publication9f6e12e0-098e-4548-ab81-75552e8385e7virtual::229-19f6e12e0-098e-4548-ab81-75552e8385e7virtual::229-1ORIGINALImplementation of Active Data Selection Algorithms for Data Choosing in ASV Systems.pdfImplementation of Active Data Selection Algorithms for Data Choosing in ASV Systems.pdfRestricción de acceso por un año, esto por si acaso existe algún evento en el que se pueda publicar el articulo.application/pdf452074https://repositorio.uniandes.edu.co/bitstreams/f93beba8-a0cc-497c-8179-c6bbe3ecc332/downloadd9921c9e9d7061b558245ddf9bec4ad9MD52autorizacion tesis.pdfautorizacion tesis.pdfHIDEapplication/pdf241954https://repositorio.uniandes.edu.co/bitstreams/f105fe38-de1b-41d6-be6e-32bab4d90fad/downloade55ff3fd2c68dc12071bdce45695ee59MD51CC-LICENSElicense_rdflicense_rdfapplication/rdf+xml; charset=utf-8908https://repositorio.uniandes.edu.co/bitstreams/1ce538e8-f5d6-4c74-8c80-f78d5a3efab2/download0175ea4a2d4caec4bbcc37e300941108MD53LICENSElicense.txtlicense.txttext/plain; charset=utf-82535https://repositorio.uniandes.edu.co/bitstreams/7bd0ff29-36ee-4bde-b126-36bc915d696f/downloadae9e573a68e7f92501b6913cc846c39fMD54TEXTImplementation of Active Data Selection Algorithms for Data Choosing in ASV Systems.pdf.txtImplementation of Active Data Selection Algorithms for Data Choosing in ASV Systems.pdf.txtExtracted texttext/plain26762https://repositorio.uniandes.edu.co/bitstreams/0cfaa3e5-6ba0-4ce8-ac70-fc7c80231c85/downloadf4ef904ad2c1be221cf2f2ca5424bac7MD55autorizacion tesis.pdf.txtautorizacion tesis.pdf.txtExtracted texttext/plain1176https://repositorio.uniandes.edu.co/bitstreams/2990c984-8a58-406d-9939-b5c2180b285d/download530d320eb9a3b695b3582e4fc981c68eMD57THUMBNAILImplementation of Active Data Selection Algorithms for Data Choosing in ASV Systems.pdf.jpgImplementation of Active Data Selection Algorithms for Data Choosing in ASV Systems.pdf.jpgGenerated Thumbnailimage/jpeg9762https://repositorio.uniandes.edu.co/bitstreams/e8a097ab-8c29-4580-b939-a49b9cf94a74/download734d141c1598fab4b0dce15627b25d96MD56autorizacion tesis.pdf.jpgautorizacion tesis.pdf.jpgGenerated Thumbnailimage/jpeg11252https://repositorio.uniandes.edu.co/bitstreams/8871c1b0-4495-42b0-bcdc-1110e2124cd7/download5811fdeb69405a32c18155d67596c071MD581992/73644oai:repositorio.uniandes.edu.co:1992/736442024-02-16 14:51:41.33http://creativecommons.org/licenses/by/4.0/Attribution 4.0 Internationalembargohttps://repositorio.uniandes.edu.coRepositorio institucional Sénecaadminrepositorio@uniandes.edu.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 |