Methods for Phonetic Scraping of Youtube Videos

Adrien Méli (1) , Steven Coats (2) , Nicolas Ballier (1, 3)

1 CLILLAC-ARP (URP_3967) - Centre de Linguistique Inter-langues, de Lexicologie, de Linguistique Anglaise et de Corpus
2 University of Oulu
3 LLF - UMR7110 - Laboratoire de Linguistique Formelle

This paper discusses two pipelines for the auto- matic collection of automatic speech recognition (ASR) transcripts and audio content from YouTube videos and subsequent phonetic analysis: PEASYV (Phonetic Extraction and Alignment of Subtitled YouTube Videos) and YTPP (YouTube Phonetics Pipeline). The pipelines differ somewhat in terms of processing steps as well as the tools used for forced alignment, but produce comparable results. The two pipelines may be useful for large-scale collection of acoustic data for phonetic analysis.

Mots clés: ASR Automatic Speech Recognition; Forced alignment; Big data; Phonetic analysis

Domaines: Intelligence artificielle [cs.AI]; Linguistique

Adrien Méli, Steven Coats, Nicolas Ballier. Methods for Phonetic Scraping of Youtube Videos. 6th International Conference on Natural Language and Speech Processing (ICNLSP 2023), Mourad Abbas, Abed Alhakim Freihat, Dec 2023, Trento (Italy), France. pp.244-249. ⟨hal-04547365⟩