Modelling of multi-modal data in LiRI Corpus Platform and beyond

While there exist common solutions and standard practices to model text-only corpora, multimodal conversation corpora present the unique challenge of integrating textual transcripts, temporal information and annotations that can equally relate to speech and non-speech modalities. The LiRI Corpus Platform (LCP) was designed to accommodate both textual and multimodal corpora, allowing for the modeling of time-aligned annotations associated with multimedia files, making it possible for users of the platform to efficiently browse and query multimodal corpora. LCP aims to become a solution of reference for hosting multimodal corpora and we especially encourage curators of such corpora to attend this session.

In this session, we report on the process of importing a multimodal conversation corpus into LCP as part of the FAIR FI-LD project. We will give an overview of the different file formats that were involved in the modelling of the corpus and explain how we converged on using the TEI/ISO format as our interoperable standard. Participants  engage in a practical exercise as they are guided through a simple Python script to import a simplified sample corpus into LCP. We end the session with a general discussion about setting up workflows for the curation of multi-modal corpora and about the possibilities of automatizing the extraction and annotation of speech from videos.

The CLARIN-CH Training Sessions 2025: Exploring Swiss Language Resources and Tools

This training series took place during the spring semester 2025 and was organized by members of the CLARIN-CH ecosystem of infrastructures with the aim to introduce and deepen participants' knowledge of the Swiss national infrastructure for language data.