Publication: Herramienta de transcripción automática de los comunicados al pasaje aéreo
Loading...
Identifiers
Publication date
2014-03
Defense date
2014-03-21
Authors
Tutors
Journal Title
Journal ISSN
Volume Title
Publisher
Impact
Export
Abstract
Este proyecto consiste en conseguir la integración y autonomía de las personas con discapacidad
auditiva en la industria aeronáutica mediante la realización de una herramienta de
subtitulado de comunicados de la tripulación al pasaje aéreo. Se asume la ausencia de aplicaciones
que faciliten esta comunicación, y por ello se desarrolla una nueva herramienta. El
objetivo es transcribir y mostrar los mensajes orales a las personas con discapacidad sensorial,
en particular, auditiva.
Para el diseño y desarrollo se tienen en cuenta las limitaciones del escenario. En primer lugar,
el ruido procedente de la aeronave, que contamina el audio. En segundo lugar, el hecho de
que la locución sea habla continua en varios idiomas, en concreto inglés y español. Por último,
pero no menos importante, la transcripción se muestra como un subtítulo, semejante al utilizado
por la industria televisiva y cinematográfica. El punto de partida es un motor de reconocimiento
comercial (Dragon NaturallySpeaking) y se ha realizado, por una parte, su integración en la
herramienta y por otra, la adaptación de sus modelos a las limitaciones anteriormente mencionadas.
Se han realizado un conjunto de pruebas para simular la locución en el interior de la cabina.
Para ello se ha transcrito un audio-libro en inglés, mezclado a distintos niveles de ruido, y se
han obtenido las tasas de acierto de palabra. Los resultados sirven para mostrar la capacidad de
la herramienta a la hora de transcribir habla continua cuando es adaptada a las necesidades de
la aplicación.
The aim of this project is to achieve the integration and autonomy of people with hearing disabilities in the aviation industry by implementing a system which provides captions of the crew’s announcement to airplane passengers. Assuming the absence of applications that help with this communication, a new tool is developed. The aim is to transcribe and show the speech to people with sensory disabilities, in particular, with hearing problems. For the design and development of the tool the limitations of the environment are taken into account. First of all, the noise from the aircraft, which degrades the audio quality. Secondly, the fact that it deals with continous speech in two different languages, English and Spanish. And last but not least, the captions must be shown as a subtitle, similar to the ones used by the film and television industry. We depart from a comercial speech recognition engine (Dragon NaturallySpeaking) we have integrated into our application, on the one hand and adapted to the previously mentioned constraints, on the other. A set of objective tests was run to simulate the speech inside of an airplane. For this purpose, an English audiobook, mixed with different noise levels, was transcribed and the word accuracy rate obtained. These results profile the performance of the tool when transcribing continuous speech with the particular needs of the application.
The aim of this project is to achieve the integration and autonomy of people with hearing disabilities in the aviation industry by implementing a system which provides captions of the crew’s announcement to airplane passengers. Assuming the absence of applications that help with this communication, a new tool is developed. The aim is to transcribe and show the speech to people with sensory disabilities, in particular, with hearing problems. For the design and development of the tool the limitations of the environment are taken into account. First of all, the noise from the aircraft, which degrades the audio quality. Secondly, the fact that it deals with continous speech in two different languages, English and Spanish. And last but not least, the captions must be shown as a subtitle, similar to the ones used by the film and television industry. We depart from a comercial speech recognition engine (Dragon NaturallySpeaking) we have integrated into our application, on the one hand and adapted to the previously mentioned constraints, on the other. A set of objective tests was run to simulate the speech inside of an airplane. For this purpose, an English audiobook, mixed with different noise levels, was transcribed and the word accuracy rate obtained. These results profile the performance of the tool when transcribing continuous speech with the particular needs of the application.
Description
Keywords
Accesibilidad, Discapacidad auditiva, Reconocimiento automático del habla, RAH, Subtitulado, Aeronaves, Ruido, Página web, Robustez, Adaptación