The Artifical Intelligence and Data Initiative (AIDI) and EuroVOX co-organized a hands-on micro-workshop on AI, Data and Language on 21 June 2021. This workshop included in-depth demos and tutorials focusing on AI/data-driven language-related tools, from speech-to-text (STT) to synthetic voices.
Ben Poor, EBU project manager in Radio at the Technology & Innovation department, presented the latest developments in EuroVOX, the EBU’s open ecosystem for transcription and translation of PSM audio-visual content.
Nicholas Kennedy, Graduate R&D Engineer at BBC, gave a brief recap of the history of speech synthesis, user perceptions, and control methods, described a study about the perceptive quality of synthetic speech radio in the Synthetic Radio project, as well as challenged participants to write Speech Synthesis Markup Language (SSML).
Peggy van der Kreeft, Deutsche Welle project manager focusing on language technologies, presented the EU H2020 GoURMET project and related efforts. GoURMET develops machine translation models for low-resourced languages and domains. She outlined the goals and challenges of providing media companies with new translation models for languages that have lower coverage and therefore less training data. After a look at the models developed so far, she demonstrated how these are currently used and made available. She was then joined by her colleague Hala Attig, Innovation manager at Deutsche Welle, who described their benchmarking process.
Alexandre Rouxel, Senior project engineer in AI and Data at the EBU’s Technology & Innovation, then gave a demo of the EBU’s STT Benchmarking tool, an open-source tool for benchmarking machine learning-based STT systems, developed with Members at the EBU. This tool can be integrated into production workflows and is currently used by Deutsche Welle in the GoURMET project.
This microworkshop was for EBU Members only.
Register by: 21 Jun 2021