Using a Knowledge Base to Automatically Annotate Speech Corpora and to Identify Sociolinguistic Variation

Proceedings of the 13th Conference on Language Resources and Evaluation (LREC 2022), pages 1054–1060 Marseille, 20-25 June 2022 © European Language Resources Association (ELRA), licensed under CC-BY-NC-4.0 1054 Using a Knowledge Base to Automatically Annotate Speech Corpora and to Identify Sociolinguistic Variation Yaru Wu1,3,4, Fabian Suchanek2, Ioana Vasilescu3, Lori Lamel3, Martine Adda-Decker4 1CRISCO/EA4255, Universit´e de Caen Normandie, 14000 Caen, France, 2T´el´ecom Paris, Institut Polytechnique de Paris, France, 3LISN, Univ.Paris-Saclay, 91405 Orsay cedex, France, 4Laboratoire de Phon´etique et Phonologie (UMR7018, CNRS-Sorbonne Nouvelle), France yaru.wu@unicaen.fr, suchanek@telecom-paris.fr, {ioana.vasilescu, lori.lamel}@lisn.upsaclay.fr, martine.adda-decker@sorbonne-nouvelle.fr Abstract Speech characteristics vary from speaker to speaker. While some variation phenomena are due to the overall communication setting, others are due to diastratic factors such as gender, provenance, age, and social background. The analysis of these factors, although relevant for both linguistic and speech technology communities, is hampered by the need to annotate existing corpora or to recruit, categorise, and record volunteers as a function of targeted profiles. This paper presents a methodology that uses a knowledge base to provide speaker-specific information. This can facilitate the enrichment of existing corpora with new annotations extracted from the knowledge base. The method also helps the large scale analysis by automatically extracting instances of speech variation to correlate with diastratic features. We apply our method to an over 120-hour corpus of broadcast speech in French and investigate variation patterns linked to reduction phenomena and/or specific to connected speech such as disfluencies. We find significant differences in speech rate, the use of filler words, and the rate of non-canonical realisations of frequent segments as a function of different professional categories and age groups. Keywords: diastratic variation, knowledge base, socio-linguistic factors, journalistic French, large corpora, non-canonical realisations, automatic alignment 1. Introduction Speech characteristics depend on many factors. While some variations are due to the communicative context (e.g., formal vs familiar), others are due to individual traits of the speaker such as gender, origin, age, and socio-economic status, or to the emotional state of a speaker in a particular situation. How speech varies according to the speakers’ char- acteristics has been a key focus in variationist studies over the last fifty years, starting with Labov’s seminal work (Labov et al., 2013; Labov and Rosenfelder, 2011). More recently, speech technology has also taken on the subject, as speech-based applications can be highly impacted by uncontrolled variation (Benzeghiba et al., 2007; Goldwater et al., 2010; Adda-Decker and Lamel, 2017). Speaker-related patterns of variation are thus of interest for linguists and speech technolo- gists alike, for example to improve the performance of speech-to-text transcription systems, of speaker diarization and/or identification algorithms, of emotion recognition systems, or to customise the output of speech generation systems. However, a principled analysis of such patterns relies on the availability of speech corpora with annotations for speaker identities and profiles, or on the recruitment of volunteers, obtaining their profiles, and acquiring speech corpora from them in various communicative settings. Both are time and money consuming endeavours, and in terms of methodology tributary to the Observer’s paradox …

Content truncated to 500 words