Title
Human versus machine. Semi-automatization of delay calculations in live subtitling
Conference name
9th International conference Media for all
City
Country
Spain
Date
27/01/2021-29/01/2021
Abstract
After some time and reiterated requests from the Deaf and Hard-of-Hearing community, quality is now one of the main topics of discussion in live subtitling. From a research standpoint, it has been embraced as a necessary object of study, which has led to a recent increase in scholarly works aimed at mapping the state of the art of live subtitling in several countries. Together with accuracy, one of the technical parameters that tends to be mentioned in official guidelines and regulations dealing with quality is live subtitling delay, which should be reduced to the minimum possible extent. In this context, both the subtitling industry and researchers conducting quality studies need to come up with accurate estimations for the latency featured in their live subtitling samples. However, performing these calculations is highly time-consuming since it involves properly synchronizing the delayed subtitles contained in hours of audiovisual materials.
This presentation will deal with the semi-automatization of delay calculations in quality studies. It will begin by explaining the different methodologies that have been used to estimate live subtitling latency for large numbers of samples in several quality assessment projects undertaken in the UK, Spain, and the US. It will discuss the main advantages and drawbacks in each case, as well as the time invested to perform the delay analyses. Then, it will describe an alternative semiautomatized method based on the use of forced-alignment technologies, which automatically align orthographic transcriptions of texts with audio recordings. The potential application of these tools to systematize and optimize latency calculations will be discussed, drawing on the findings of a research project undertaken in the US which explored the closed captioning delay of 30 programs broadcast in Spanish. The results obtained point at forced-alignment tools as promising aids that may reduce the time invested in latency calculations, although human intervention is always needed, and optimal performance is only achieved in very controlled conditions.
This presentation will deal with the semi-automatization of delay calculations in quality studies. It will begin by explaining the different methodologies that have been used to estimate live subtitling latency for large numbers of samples in several quality assessment projects undertaken in the UK, Spain, and the US. It will discuss the main advantages and drawbacks in each case, as well as the time invested to perform the delay analyses. Then, it will describe an alternative semiautomatized method based on the use of forced-alignment technologies, which automatically align orthographic transcriptions of texts with audio recordings. The potential application of these tools to systematize and optimize latency calculations will be discussed, drawing on the findings of a research project undertaken in the US which explored the closed captioning delay of 30 programs broadcast in Spanish. The results obtained point at forced-alignment tools as promising aids that may reduce the time invested in latency calculations, although human intervention is always needed, and optimal performance is only achieved in very controlled conditions.