Dubious dubbing?: Análisis del uso de la inteligencia artificial para el doblaje de redes sociales en la plataforma YouTube.
Loading...
Date
2025
Journal Title
Journal ISSN
Volume Title
Publisher
Universidad de Concepción
Abstract
Esta investigación tiene como objetivo analizar la calidad del doblaje generado con inteligencia artificial, utilizando la plataforma HeyGen, en comparación con el doblaje tradicional. Para ello, se seleccionó como corpus un video originalmente en inglés de un pódcast publicado en YouTube por Tom Bilyeu, junto con su versión doblada al español de su canal Tom Bilyeu Español y se generó una versión doblada con IA mediante HeyGen. Para definir el corpus utilizado, se consideraron los primeros 15 minutos de ambos videos y la versión original en inglés para garantizar un contraste. El análisis se basó en el modelo TP propuesto por Spiteri Miggiani (2023) para medir la calidad de un doblaje. Este permitió identificar y clasificar errores en seis parámetros de calidad, los cuales a su vez poseían su propia clasificación específica de errores. Los resultados mostraron un total de 99 errores en el doblaje tradicional y 97 en el doblaje con IA. Sin embargo, la distribución de errores fue diferente en cada modalidad. En el doblaje tradicional predominaron los errores de sincronización labial con 46, mientras que, en el doblaje con IA, los más frecuentes fueron los de naturalidad en los diálogos con 46 errores. A partir de estos hallazgos se podría concluir que, si bien la cantidad total de errores fue similar, el doblaje con IA presentó limitaciones relevantes en parámetros que afectan directamente la comprensión y credibilidad del mensaje, a diferencia del doblaje tradicional, que tuvo un mejor desempeño en este aspecto.
The aim of this research is to analyze the quality of dubbing generated by AI using the HeyGen platform, in comparison with traditional dubbing. The corpus comprises a podcast video originally in English published on YouTube by Tom Bilyeu, along with its dubbed Spanish version from the channel Tom Bilyeu Español and together with an AI-dubbed version that was generated by HeyGen. In order to define the sample, the first 15 minutes of the English original video and both Spanish versions were selected to ensure contrast. The analysis was based on the Spiteri Miggiani’s (2023) TP model to measure the quality of dubbing, which enabled the identification and classification of errors across six quality parameters, each one with its own error taxonomy. Results indicate a total of 99 errors in the traditional dubbing and 97 in the AI-generated dubbing, however, the distribution of errors was different in each one of them. In traditional dubbing, adequate lip-synchronisation errors predominated, with a total of 46 errors, whereas in the AI-generated dubbing, the most frequent errors were those related to natural-sounding language with a total of 46 errors. These findings suggest that, although the overall number of errors was similar, the AI-generated dubbing had relevant limitations in parameters that directly affect the comprehension and credibility of the message, whereas the traditional dubbing performed better on this regard.
The aim of this research is to analyze the quality of dubbing generated by AI using the HeyGen platform, in comparison with traditional dubbing. The corpus comprises a podcast video originally in English published on YouTube by Tom Bilyeu, along with its dubbed Spanish version from the channel Tom Bilyeu Español and together with an AI-dubbed version that was generated by HeyGen. In order to define the sample, the first 15 minutes of the English original video and both Spanish versions were selected to ensure contrast. The analysis was based on the Spiteri Miggiani’s (2023) TP model to measure the quality of dubbing, which enabled the identification and classification of errors across six quality parameters, each one with its own error taxonomy. Results indicate a total of 99 errors in the traditional dubbing and 97 in the AI-generated dubbing, however, the distribution of errors was different in each one of them. In traditional dubbing, adequate lip-synchronisation errors predominated, with a total of 46 errors, whereas in the AI-generated dubbing, the most frequent errors were those related to natural-sounding language with a total of 46 errors. These findings suggest that, although the overall number of errors was similar, the AI-generated dubbing had relevant limitations in parameters that directly affect the comprehension and credibility of the message, whereas the traditional dubbing performed better on this regard.
Description
Tesis presentada para optar al grado de Licenciado en Traductología.
Keywords
Traducción e interpretación, Inteligencia artificial, Podcasting