FCSIT Researchers at the Jelinek Summer Workshop on Speech and Language Technology 2023, France The annual Jelinek Summer Workshop on Speech and Language Technology (JSALT) 2023, organized by John Hopkins University’s Center for Language and Speech Processing, currently running on its 30th edition in France this year, saw a team of four researchers from the Faculty of Computer Science and Information Technology, UNIMAS, as part its international team of participants chosen from a highly-competitive selection process. The UNIMAS team, consisting of Dr Sarah Flora Samson Juan, Dr Suhaila Saee, Mohd Zulhafiz Rahim and Muhammad Abdullah Yusof, spent 6 weeks from 26 June to 4 August in the city of Le Mans in northwestern France to undergo this intensive research program, working closely with experts and fellow specialists in the field of speech and natural language processing to tackle the field’s challenges that have not yet been thoroughly investigated. At JSALT2023 the UNIMAS team worked on the topic of explainability for diarization, focusing on Sarawak Malaysia language data. The Sarawak Malay language is considered an under resourced language, in which a language has limited amount of linguistic data, resources, and tools. Diarization is a the process of partitioning an audio stream that may contain human speech, music or any background sound. Over the past five years, scientists have developed methods to perform automatic partitioning through deep learning methods. Explainability for diarization is proposed in the workshop to answer several research questions, such as, “Who spoke when?”, and “Why is there a speaker boundary in that audio segment?” The JSALT2023 program was a demanding one, featuring daily progress meetings, weekly progress presentations, and weekly seminars that culminated in the final presentation of the solutions produced after 6 weeks of intensive research and development. Besides the topic of explainability for diarization, other JSALT2023 participants worked on topics of text and context embeddings, finite state methods with modern neural architectures for speech applications and beyond, and automatic design of conversational models from observation of human-to-human conversation. Speaking about the experience in France, postgraduate students Mohd Zulhafiz Rahim and Muhammad Abdullah Yusof stated that JSALT2023 has provided them deeper understanding and valuable insights especially in diarization, the immersive experience of working directly with experts, the opportunity to explore new frontiers in the field, especially in under resourced languages such as Sarawak Malay language, and the knowledge of various tools and methods in natural language processing. The journey to JSALT2023 began in October 2022 for the team, in which all prospective participants submitted proposals which were then selected through an interactive peer review process, followed by presentation and debate. JSALT2023 is also co-hosted by the LIUM (Laboratoire Informatique de l'Université du Mans) and Allo Media. The opportunity to travel to JSALT2023 for the UNIMAS team was made possible through the ESPERANTO Consortium program supported by European Union’s Horizon 2020 research and innovation program under the Marie Skłodowska-Curie grant. UNIMAS is one of the 19 partners involved in the consortium to collaboratively conduct research in the field of speech processing applications.
ESPERANTO Research Consortium comprises 19 partners from various institutions and companies from four continents: Europe, Africa, Asia and America.
Presentation by Prof Sanjeev Khudanpur from John Hopkins University, USA.