Publicação

ArtiFlow: An Integrated Pipeline for Identifying and Analyzing Speech Disfluencies: A Deep Learning Framework for Enhanced Transcription, Classification and Interpretation of Disfluent Speech

Ver documento

Detalhes bibliográficos
Resumo:Automatic Speech Recognition (ASR) systems struggle with speech disfluencies like repetitions and filled pauses, which are common in spontaneous speech and disorders such as stuttering. This performance gap limits the accessibility and clinical utility of ASR technology. This research proposes a comprehensive Deep Learning framework to improve the transcription, classification, and interpretation of disfluent speech. The multi-stage methodology begins by fine-tuning state-of-the-art ASR models on the FluencyBank corpus to accurately transcribe disfluent events. Next, a custom-trained ModernBERT model performs token-level classification to identify and label specific disfluencies in the transcript. Finally, a specialized Large Language Model (LLM) provides a structured, context-aware analysis of the identified patterns. The integrated pipeline is demonstrated through a functional prototype, ArtiFlow (Articulate Flow), which showcases the system's end-to-end capabilities. Empirical results validate the framework's effectiveness. After a comparative analysis of different ASR model families, the Whisper Large v3 Turbo model was selected, offering an optimal balance between high transcription accuracy (12.94% Word Error Rate) and enhanced inference speed, while leveraging a stable and accessible implementation framework. The subsequent disfluency classifier performs with high reliability (weighted F1-score of 0.9512) and the selected LLM demonstrates strong capabilities in generating structured analyses. This thesis establishes a robust, modular system for both identifying and interpreting speech disfluencies, providing a foundation for advanced support tools. The work contributes to speech processing and assistive technology, offering significant potential for applications in clinical diagnosis, speech therapy, and the development of more inclusive human-computer interaction systems.
Autores principais:Pérez, Ariel Enrique Cerda
Assunto:Automatic Speech Recognition Speech Disfluencies Large Language Models Spoken Language Processing Stuttering SDG 3 - Good health and well-being
Ano:2025
País:Portugal
Tipo de documento:dissertação de mestrado
Tipo de acesso:acesso embargado
Instituição associada:Universidade Nova de Lisboa
Idioma:inglês
Origem:Repositório Institucional da UNL
_version_ 1868982710970613760
author Pérez, Ariel Enrique Cerda
author_facet Pérez, Ariel Enrique Cerda
author_role author
contributor_name_str_mv Bação, Fernando José Ferreira Lucas
RUN
country_str PT
creators_json_txt [{\"Person.name\":\"Pérez, Ariel Enrique Cerda\"}]
datacite.contributors.contributor.contributorName.fl_str_mv Bação, Fernando José Ferreira Lucas
RUN
datacite.creators.creator.creatorName.fl_str_mv Pérez, Ariel Enrique Cerda
datacite.date.Accepted.fl_str_mv 2025-10-28T00:00:00Z
datacite.date.available.fl_str_mv 2025-11-07T15:55:19Z
datacite.date.embargoed.fl_str_mv 2025-11-07T15:55:19Z
datacite.rights.fl_str_mv http://purl.org/coar/access_right/c_f1cf
datacite.subjects.subject.fl_str_mv Automatic Speech Recognition
Speech Disfluencies
Large Language Models
Spoken Language Processing
Stuttering
SDG 3 - Good health and well-being
datacite.titles.title.fl_str_mv ArtiFlow: An Integrated Pipeline for Identifying and Analyzing Speech Disfluencies: A Deep Learning Framework for Enhanced Transcription, Classification and Interpretation of Disfluent Speech
dc.contributor.none.fl_str_mv Bação, Fernando José Ferreira Lucas
RUN
dc.creator.none.fl_str_mv Pérez, Ariel Enrique Cerda
dc.date.Accepted.fl_str_mv 2025-10-28T00:00:00Z
dc.date.available.fl_str_mv 2025-11-07T15:55:19Z
dc.date.embargoed.fl_str_mv 2025-11-07T15:55:19Z
dc.format.none.fl_str_mv application/pdf
dc.identifier.none.fl_str_mv http://hdl.handle.net/10362/190288
dc.language.none.fl_str_mv eng
dc.rights.cclincense.fl_str_mv http://creativecommons.org/licenses/by/4.0/
dc.rights.none.fl_str_mv http://purl.org/coar/access_right/c_f1cf
dc.subject.none.fl_str_mv Automatic Speech Recognition
Speech Disfluencies
Large Language Models
Spoken Language Processing
Stuttering
SDG 3 - Good health and well-being
dc.title.fl_str_mv ArtiFlow: An Integrated Pipeline for Identifying and Analyzing Speech Disfluencies: A Deep Learning Framework for Enhanced Transcription, Classification and Interpretation of Disfluent Speech
dc.type.none.fl_str_mv http://purl.org/coar/resource_type/c_bdcc
description Automatic Speech Recognition (ASR) systems struggle with speech disfluencies like repetitions and filled pauses, which are common in spontaneous speech and disorders such as stuttering. This performance gap limits the accessibility and clinical utility of ASR technology. This research proposes a comprehensive Deep Learning framework to improve the transcription, classification, and interpretation of disfluent speech. The multi-stage methodology begins by fine-tuning state-of-the-art ASR models on the FluencyBank corpus to accurately transcribe disfluent events. Next, a custom-trained ModernBERT model performs token-level classification to identify and label specific disfluencies in the transcript. Finally, a specialized Large Language Model (LLM) provides a structured, context-aware analysis of the identified patterns. The integrated pipeline is demonstrated through a functional prototype, ArtiFlow (Articulate Flow), which showcases the system's end-to-end capabilities. Empirical results validate the framework's effectiveness. After a comparative analysis of different ASR model families, the Whisper Large v3 Turbo model was selected, offering an optimal balance between high transcription accuracy (12.94% Word Error Rate) and enhanced inference speed, while leveraging a stable and accessible implementation framework. The subsequent disfluency classifier performs with high reliability (weighted F1-score of 0.9512) and the selected LLM demonstrates strong capabilities in generating structured analyses. This thesis establishes a robust, modular system for both identifying and interpreting speech disfluencies, providing a foundation for advanced support tools. The work contributes to speech processing and assistive technology, offering significant potential for applications in clinical diagnosis, speech therapy, and the development of more inclusive human-computer interaction systems.
dirty 0
eu_rights_str_mv embargoedAccess
format masterThesis
fulltext.url.fl_str_mv https://run.unl.pt/bitstreams/cd1628a4-c00b-4f8f-928a-8805b90129aa/download
id run_a8db9f919f5e4e3d3d9b00a47ad691f0
identifier.url.fl_str_mv http://hdl.handle.net/10362/190288
inst_facet_str urn:organizationAcronym:unl{{{_:::_}}}Universidade Nova de Lisboa
instacron_str unl
institution Universidade Nova de Lisboa
instname_str Universidade Nova de Lisboa
language eng
network_acronym_str run
network_name_str Repositório Institucional da UNL
oai_identifier_str oai:run.unl.pt:10362/190288
organization_str_mv urn:organizationAcronym:unl
person_str_mv Pérez, Ariel Enrique Cerda
publishDate 2025
repo_facet_str urn:repositoryAcronym:run{{{_:::_}}}Repositório Institucional da UNL
reponame_str Repositório Institucional da UNL
repository_id_str urn:repositoryAcronym:run
service_str_mv urn:repositoryAcronym:run
spelling engpt_PTAutomatic Speech Recognition (ASR) systems struggle with speech disfluencies like repetitions and filled pauses, which are common in spontaneous speech and disorders such as stuttering. This performance gap limits the accessibility and clinical utility of ASR technology. This research proposes a comprehensive Deep Learning framework to improve the transcription, classification, and interpretation of disfluent speech. The multi-stage methodology begins by fine-tuning state-of-the-art ASR models on the FluencyBank corpus to accurately transcribe disfluent events. Next, a custom-trained ModernBERT model performs token-level classification to identify and label specific disfluencies in the transcript. Finally, a specialized Large Language Model (LLM) provides a structured, context-aware analysis of the identified patterns. The integrated pipeline is demonstrated through a functional prototype, ArtiFlow (Articulate Flow), which showcases the system's end-to-end capabilities. Empirical results validate the framework's effectiveness. After a comparative analysis of different ASR model families, the Whisper Large v3 Turbo model was selected, offering an optimal balance between high transcription accuracy (12.94% Word Error Rate) and enhanced inference speed, while leveraging a stable and accessible implementation framework. The subsequent disfluency classifier performs with high reliability (weighted F1-score of 0.9512) and the selected LLM demonstrates strong capabilities in generating structured analyses. This thesis establishes a robust, modular system for both identifying and interpreting speech disfluencies, providing a foundation for advanced support tools. The work contributes to speech processing and assistive technology, offering significant potential for applications in clinical diagnosis, speech therapy, and the development of more inclusive human-computer interaction systems.application/pdfpt_PTArtiFlow: An Integrated Pipeline for Identifying and Analyzing Speech Disfluencies: A Deep Learning Framework for Enhanced Transcription, Classification and Interpretation of Disfluent SpeechPérez, Ariel Enrique CerdaBação, Fernando José Ferreira LucasHostingInstitutionOrganizationalRUNe-mailmailto:run@unl.ptrun@unl.ptURNurn:tid:2040714102025-11-07T15:55:19Z2025-10-282025-10-28T00:00:00ZHandlehttp://hdl.handle.net/10362/190288http://purl.org/coar/access_right/c_f1cfembargoed accessAutomatic Speech RecognitionSpeech DisfluenciesLarge Language ModelsSpoken Language ProcessingStutteringSDG 3 - Good health and well-being2659439 bytesliteraturehttp://purl.org/coar/resource_type/c_bdccmaster thesis2025-10-28http://creativecommons.org/licenses/by/4.0/http://purl.org/coar/access_right/c_f1cfapplication/pdffulltexthttps://run.unl.pt/bitstreams/cd1628a4-c00b-4f8f-928a-8805b90129aa/download
spellingShingle ArtiFlow: An Integrated Pipeline for Identifying and Analyzing Speech Disfluencies: A Deep Learning Framework for Enhanced Transcription, Classification and Interpretation of Disfluent Speech
Pérez, Ariel Enrique Cerda
Automatic Speech Recognition
Speech Disfluencies
Large Language Models
Spoken Language Processing
Stuttering
SDG 3 - Good health and well-being
status SINGLETON
subject.fl_str_mv Automatic Speech Recognition
Speech Disfluencies
Large Language Models
Spoken Language Processing
Stuttering
SDG 3 - Good health and well-being
title ArtiFlow: An Integrated Pipeline for Identifying and Analyzing Speech Disfluencies: A Deep Learning Framework for Enhanced Transcription, Classification and Interpretation of Disfluent Speech
title_full ArtiFlow: An Integrated Pipeline for Identifying and Analyzing Speech Disfluencies: A Deep Learning Framework for Enhanced Transcription, Classification and Interpretation of Disfluent Speech
title_fullStr ArtiFlow: An Integrated Pipeline for Identifying and Analyzing Speech Disfluencies: A Deep Learning Framework for Enhanced Transcription, Classification and Interpretation of Disfluent Speech
title_full_unstemmed ArtiFlow: An Integrated Pipeline for Identifying and Analyzing Speech Disfluencies: A Deep Learning Framework for Enhanced Transcription, Classification and Interpretation of Disfluent Speech
title_short ArtiFlow: An Integrated Pipeline for Identifying and Analyzing Speech Disfluencies: A Deep Learning Framework for Enhanced Transcription, Classification and Interpretation of Disfluent Speech
title_sort ArtiFlow: An Integrated Pipeline for Identifying and Analyzing Speech Disfluencies: A Deep Learning Framework for Enhanced Transcription, Classification and Interpretation of Disfluent Speech
topic Automatic Speech Recognition
Speech Disfluencies
Large Language Models
Spoken Language Processing
Stuttering
SDG 3 - Good health and well-being
topic_facet Automatic Speech Recognition
Speech Disfluencies
Large Language Models
Spoken Language Processing
Stuttering
SDG 3 - Good health and well-being
url http://hdl.handle.net/10362/190288
visible 1