Developing Resources for Automated Speech Processing of Quebec French

Proceedings of the 12th Conference on Language Resources and Evaluation (LREC 2020), pages 5323–5328 Marseille, 11–16 May 2020 c⃝ European Language Resources Association (ELRA), licensed under CC-BY-NC 5323 Developing Resources for Automated Speech Processing of Quebec French Mélanie Lancien1,2, Marie-Hélène Côté1, Brigitte Bigi3 Université de Lausanne1, Université du Québec à Chicoutimi2, LPL-CNRS-Aix Marseille Univ3 Lausanne (Suisse), Chicoutimi (Canada), Aix-en-Provence (France) melanie.lancien@unil.ch, marie-helene.cote@unil.ch, brigitte.bigi@lpl-aix.fr Abstract The analysis of the structure of speech nearly always rests on the alignment of the speech recording with a phonetic transcription. Nowadays several tools can perform this speech segmentation automatically. However, none of them carries out the automatic segmentation of Quebec French (QF hereafter) in a proper way. Contrary to what could be assumed, the acoustics and phonotactics of QF differs widely from that of France French (FF hereafter). To adequately segment QF, features like diphthongization of long vowels and affrication of coronal stops have to be taken into account. Thus acoustic models for automatic segmentation must be trained on speech samples exhibiting those phenomena. Dictionaries and lexicons must also be adapted and integrate differences in lexical units (such as very frequent words in QF that are not used in FF) and in the phonology of QF (such as the existence of tense and lax high vowels in QF but not in FF). This paper presents the development of linguistic resources to be included into the SPPAS software tool in order to get Text normalization, Phonetization, Alignment and Syllabification. We adapted the existing French lexicon and developed a QF-specific pronunciation dictionary. We then created an acoustic model from the existing ones and adapted it with 5 minutes of manually time-aligned data. These new resources are all freely distributed with SPPAS version 2.7; they perform the full process of speech segmentation in Quebec French. Keywords: Speech segmentation, Quebec French, resources, dictionary, phonetics 1. Introduction The physical characteristics of speech can be observed in the dynamics of the production of sounds. One of the main goals of phonetic studies is to capture, describe and explain the physical characteristics of linguistic units. First, such studies mostly consist in determining the relevant units and aligning them with segments of the audio signal. Then they measure parameters of the units (e.g. their duration) and, finally, they perform some descriptive statistical anal- yses. Annotating the units is thus crucially important. In this context, the required annotations consist in perform- ing a phonetic alignment, i.e. labelling and segmenting phonemes. Labelling determines the sounds of a transcrip- tion of a speech recording. Segmenting consists in pairing these sounds with time-stamps to indicate the localization and thus the length of the sounds in the speech recording. Linguists need tools for the analysis of these features of speech sounds. The most popular software tool to perform these annotations manually is Praat (Boersma and Weenink, 2018), but other tools combine ease of use with a broad range of features like Phonedit (Teston et al., 1999) or An- notationPro (Klessa et al., 2013). Several tools for automatic phonetization …

Content truncated to 500 words