Desarrollo de una herramienta computacional para la evaluación de la calidad de voz con base en la escala GRBAS
Voice quality analysis has become a routine activity in clinics and hospitals, where they are performed by voice professionals (speech therapists); These analyses are generally performed based on the GRBAS scale, and present subjective characteristics highly influenced by experience, level of educat...
Saved in:
| Main Authors: | , |
|---|---|
| Other Authors: | |
| Format: | Trabajo de grado (Pregrado y/o Especialización) |
| Language: | spa |
| Published: |
Universidad Antonio Nariño
2022
|
| Subjects: | |
| Online Access: | http://repositorio.uan.edu.co/handle/123456789/5963 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| _version_ | 1812648477950214144 |
|---|---|
| author | Ruano Bolaños, Jesús Andrés Alegría Cardona, Cristian David |
| author2 | Villamarín Muñoz, Julián Antonio |
| author_facet | Villamarín Muñoz, Julián Antonio Ruano Bolaños, Jesús Andrés Alegría Cardona, Cristian David |
| author_sort | Ruano Bolaños, Jesús Andrés |
| collection | DSpace |
| description | Voice quality analysis has become a routine activity in clinics and hospitals, where they are
performed by voice professionals (speech therapists); These analyses are generally
performed based on the GRBAS scale, and present subjective characteristics highly
influenced by experience, level of education of staff, among others (Gordillo, 2018).
For this project, the database of synthetic voices developed by the Antonio Nariño
University was implemented through the Evaper application, with the purpose of
developing a computational tool for the evaluation of voice quality, through the extraction
of both acoustic and statistical vocal characteristics, and the implementation of machine
learning systems to give a respective diagnosis based on the GRBAS scale. As a first result,
it was found that the algorithm in charge of performing the extraction of vocal features
presented a strong level of correlation with respect to the Praat software, software that was
considered as a standard system due to its very significant trajectory in the field of speech
therapy; reaching a Spearman rho correlation > 0.85, thus validating the algorithm
dedicated to the extraction of features, and the implemented methodological process. As a
second result, it was found that the classification models implemented in this project
presented a high level of accuracy, with the exception of one of the parameters of the
female gender (Roughness), due to the existence of an error in the database, since it
presented a lack of information for this gender; the results obtained in percentage scale of
the level of accuracy of the models that make up the tool are: Hoarseness Models for male
gender =71.2% , Tension Asthenia Models for male gender = 84.8% , Tension Asthenia
Models for male gender = 93.3%, Grade Models for male gender = 93.6%, Hoarseness
Models for female gender = 40.5% , Tension Asthenia Models for female gender = 90.5% ,
Tension Asthenia Models for female gender = 97.7%, Grade Models for female gender =
95.3%. As a last result, after performing a preliminary preview of phase 2, which consisted
of evaluating the performance of the tool implemented 10 real voices provided by the
Universidad del Valle, low levels of accuracy and correlation were obtained, being these
values equal to: Accuracy <= 50% and -0,2 < Correlation <= 0,5.
The results obtained through the evaluation of the models developed with synthetic voices,
allowed validating the performance of the computational tool, however, the results after
performing the preliminary preview, despite not being very significant due to the low
amount of data, showed that it is necessary to carry out an analysis of the tool in order to
make the appropriate modifications to improve the effectiveness of its operation when
implemented with real voices. |
| format | Trabajo de grado (Pregrado y/o Especialización) |
| id | repositorio.uan.edu.co-123456789-5963 |
| institution | Repositorio Digital UAN |
| language | spa |
| publishDate | 2022 |
| publisher | Universidad Antonio Nariño |
| record_format | dspace |
| spelling | repositorio.uan.edu.co-123456789-59632024-10-09T23:22:03Z Desarrollo de una herramienta computacional para la evaluación de la calidad de voz con base en la escala GRBAS Ruano Bolaños, Jesús Andrés Alegría Cardona, Cristian David Villamarín Muñoz, Julián Antonio Subjetividad Machine learning Procesamiento de señales GRBAS Calidad vocal Herramienta computacional Subjectivity Computational tool Vocal quality GRBAS Signal processing Machine learning Voice quality analysis has become a routine activity in clinics and hospitals, where they are performed by voice professionals (speech therapists); These analyses are generally performed based on the GRBAS scale, and present subjective characteristics highly influenced by experience, level of education of staff, among others (Gordillo, 2018). For this project, the database of synthetic voices developed by the Antonio Nariño University was implemented through the Evaper application, with the purpose of developing a computational tool for the evaluation of voice quality, through the extraction of both acoustic and statistical vocal characteristics, and the implementation of machine learning systems to give a respective diagnosis based on the GRBAS scale. As a first result, it was found that the algorithm in charge of performing the extraction of vocal features presented a strong level of correlation with respect to the Praat software, software that was considered as a standard system due to its very significant trajectory in the field of speech therapy; reaching a Spearman rho correlation > 0.85, thus validating the algorithm dedicated to the extraction of features, and the implemented methodological process. As a second result, it was found that the classification models implemented in this project presented a high level of accuracy, with the exception of one of the parameters of the female gender (Roughness), due to the existence of an error in the database, since it presented a lack of information for this gender; the results obtained in percentage scale of the level of accuracy of the models that make up the tool are: Hoarseness Models for male gender =71.2% , Tension Asthenia Models for male gender = 84.8% , Tension Asthenia Models for male gender = 93.3%, Grade Models for male gender = 93.6%, Hoarseness Models for female gender = 40.5% , Tension Asthenia Models for female gender = 90.5% , Tension Asthenia Models for female gender = 97.7%, Grade Models for female gender = 95.3%. As a last result, after performing a preliminary preview of phase 2, which consisted of evaluating the performance of the tool implemented 10 real voices provided by the Universidad del Valle, low levels of accuracy and correlation were obtained, being these values equal to: Accuracy <= 50% and -0,2 < Correlation <= 0,5. The results obtained through the evaluation of the models developed with synthetic voices, allowed validating the performance of the computational tool, however, the results after performing the preliminary preview, despite not being very significant due to the low amount of data, showed that it is necessary to carry out an analysis of the tool in order to make the appropriate modifications to improve the effectiveness of its operation when implemented with real voices. El análisis de la calidad de voz se ha convertido en una actividad rutinaria en clínicas y hospitales, donde son realizadas por profesionales de la voz (logopedas); estos análisis generalmente se realizan con base en la escala GRBAS, y presentan características subjetivas altamente influenciadas por la experiencia, nivel de educación del personal, entre otros (Gordillo, 2018). Para este proyecto se implementó la base de datos de voces sintéticas desarrollada por la Universidad Antonio Nariño por medio de la aplicación Evaper, con el propósito de desarrollar una herramienta computacional para la evaluación de la calidad de voz, mediante la extracción de características vocales tanto acústicas como estadísticas, y la implementación de sistemas de machine learning para dar un respectivo diagnóstico con base en la escala GRBAS. Como primer resultado se encontró que el algoritmo encargado de realizar la extracción de las características vocales presentó un fuerte nivel de correlación con respecto al software Praat, software que se consideró como sistema estándar debido a su muy significativa trayectoria en el campo de la logopedia; llegando a obtener una correlación de Spearman rho > 0.85, validando de esta manera el algoritmo dedicado a la extracción de características y el proceso metodológico implementado. Como segundo resultado, se encontró que los modelos de clasificación implementados en este proyecto presentaron un alto nivel de exactitud, a excepción de uno de los parámetros del género femenino (Ronquera), debido a la existencia de un error en la base de datos, ya que presentaba una falta de información para este género; los resultados obtenidos en escala porcentual del nivel de exactitud de los modelos que conforman la herramienta son: Modelos de Ronquera para género masculino =71,2% , Modelos de Soplosidad para género masculino = 84,8% , Modelos de Astenia Tensión para género masculino = 93,3%, Modelos de Grado para género masculino = 93,6%, Modelos de Ronquera para género femenino = 40,5% , Modelos de Soplosidad para género femenino = 90,5% , Modelos de Astenia Tensión para género femenino = 97,7%, Modelos de Grado para género femenino = 95,3%. Como último resultado, fruto de realizar un avance preliminar de la fase 2, la cual consistió en evaluar el desempeño de la herramienta implementando 10 voces reales proporcionada por la Universidad del Valle, se obtuvieron bajos niveles de exactitud y correlación, siendo estos valores iguales a: Exactitud <= 50% y -0,2 < Correlación <= 0,5. Los resultados obtenidos mediante la evaluación de los modelos desarrollados con voces sintéticas, permitieron validar el funcionamiento de la herramienta computacional, sin embargo, los resultados tras realizar el avance preliminar, a pesar de no ser muy significativos por la baja cantidad de datos, demostraron que es necesario llevar a cabo un análisis de la herramienta con el fin de realizar las modificaciones pertinentes que mejoren la eficacia en su funcionamiento al momento de implementarla con voces reales. Ingeniero(a) Biomédico(a) Pregrado Presencial Monografía 2022-02-21T13:47:45Z 2022-02-21T13:47:45Z 2021-11-26 Trabajo de grado (Pregrado y/o Especialización) info:eu-repo/semantics/acceptedVersion http://purl.org/coar/resource_type/c_7a1f http://purl.org/coar/version/c_970fb48d4fbd8a85 http://repositorio.uan.edu.co/handle/123456789/5963 Aguilar, H., & Vélez Julia. (2016). Relación entre las pruebas aerodinámicas de la fonación con la escala GRBAS para alertar riesgo de disodeas en cantantes en formación de la Universidad del Valle, Santiago de Cali - 2016. Universidad del Valle. Barrios, J. (2019, July 26). La matriz de confusión y sus métricas. Health Big Data. Behlau, M. (2014). Voxmetria (3.3.). CTS informática. Belletti, A. (2018). Valor Cuadrático Medio o RMS. SCRIBD. Boersma, P. (2009, October 10). Should Jitter Be Measured by Peak Picking or by Waveform Matching? Folia Phoniatrica et Logopaedica Boersma, P., & Weenink, D. (2011). Praat (6.1.52). Camacho, C. (2007). COEFICIENTE DE CORRELACIÓN LINEAL DE PEARSON. In CORRELACIÓN LINEAL DE PEARSON. Centre for speech technology. (2019). WaveSurfer (1.8.8). speech.kth.se. Cerda, J., & Cifuentes, L. (2012). Uso de curvas ROC en investigación clínica. Aspectos teórico-prácticos. Revista Chilena de Infectología. Clavbo, B. (2006). Method and device for speech analysis (Patent No. US7092874). instname:Universidad Antonio Nariño reponame:Repositorio Institucional UAN repourl:https://repositorio.uan.edu.co/ spa Acceso abierto Attribution-NonCommercial-NoDerivatives 4.0 International (CC BY-NC-ND 4.0) https://creativecommons.org/licenses/by-nc-nd/4.0/ info:eu-repo/semantics/openAccess http://purl.org/coar/access_right/c_abf2 application/pdf application/pdf application/pdf Colombia (Popayán, Cauca ) Universidad Antonio Nariño Ingeniería Biomédica Facultad de Ingeniería Mecánica, Electrónica y Biomédica Popayán - Alto Cauca |
| spellingShingle | Subjetividad Machine learning Procesamiento de señales GRBAS Calidad vocal Herramienta computacional Subjectivity Computational tool Vocal quality GRBAS Signal processing Machine learning Ruano Bolaños, Jesús Andrés Alegría Cardona, Cristian David Desarrollo de una herramienta computacional para la evaluación de la calidad de voz con base en la escala GRBAS |
| title | Desarrollo de una herramienta computacional para la evaluación de la calidad de voz con base en la escala GRBAS |
| title_full | Desarrollo de una herramienta computacional para la evaluación de la calidad de voz con base en la escala GRBAS |
| title_fullStr | Desarrollo de una herramienta computacional para la evaluación de la calidad de voz con base en la escala GRBAS |
| title_full_unstemmed | Desarrollo de una herramienta computacional para la evaluación de la calidad de voz con base en la escala GRBAS |
| title_short | Desarrollo de una herramienta computacional para la evaluación de la calidad de voz con base en la escala GRBAS |
| title_sort | desarrollo de una herramienta computacional para la evaluacion de la calidad de voz con base en la escala grbas |
| topic | Subjetividad Machine learning Procesamiento de señales GRBAS Calidad vocal Herramienta computacional Subjectivity Computational tool Vocal quality GRBAS Signal processing Machine learning |
| url | http://repositorio.uan.edu.co/handle/123456789/5963 |
| work_keys_str_mv | AT ruanobolanosjesusandres desarrollodeunaherramientacomputacionalparalaevaluaciondelacalidaddevozconbaseenlaescalagrbas AT alegriacardonacristiandavid desarrollodeunaherramientacomputacionalparalaevaluaciondelacalidaddevozconbaseenlaescalagrbas |
