A Pipeline for the Creation of Multimodal Corpora from YouTube Videos (original) (raw)
2023, The 1st Workshop on Linguistic Insights from and for Multimodal Language Processing
This paper introduces an open-source pipeline for the creation of multimodal corpora from YouTube videos. It minimizes storage and bandwidth requirements, because the videos themselves need not be downloaded and can remain on YouTube’s servers. It also minimizes processing requirements by using YouTube’s automatically generated subtitles, thus avoiding a computationally expensive automatic speech recognition processing step. The pipeline combines standard tools and provides as its output a corpus file in the industry-standard vertical format used by many corpus managers. It is straightforwardly extensible with the addition of further levels of annotation and can be adapted to languages other than English.
Loading Preview
Sorry, preview is currently unavailable. You can download the paper by clicking the button above.