Multilingual automatic speech recognition (ASR) in the medical domain serves as a foundational task for various downstream applications such as speech translation, spoken language understanding, and voice-activated assistants. This technology improves patient care by enabling efficient communication across language barriers, alleviating specialized workforce shortages, and facilitating improved diagnosis and treatment, particularly during pandemics. In this work, we introduce MultiMed, the first multilingual medical ASR dataset, along with the first collection of small-to-large end-to-end medical ASR models, spanning five languages: Vietnamese, English, German, French, and Mandarin Chinese. To our best knowledge, MultiMed stands as the world’s largest medical ASR dataset across all major benchmarks: total duration, number of recording conditions, number of accents, and number of speaking roles. Furthermore, we present the first multilinguality study for medical ASR, which includes reproducible empirical baselines, a monolinguality-multilinguality analysis, Attention Encoder Decoder (AED) vs Hybrid comparative study and a linguistic analysis. We present practical ASR end-to-end training schemes optimized for a fixed number of trainable parameters that are common in industry settings. All code, data, and models are available online.

We describe our systems participated in the BioLaySumm 2023 Task 1, which aims at automatically generating lay summaries of scientific articles in a simplified way so that its content becomes easier to comprehend for non-expert readers. Our approaches are based on selecting key information by both explicit and implicit strategies. For explicit selection strategies, we conduct extractive summarization based on selecting key sentences for training abstractive summarization models. For implicit selection strategies, we utilize a method based on a factorized energy-based model, which is able to extract important information from long documents to generate summaries and achieve promising results. We build our systems using sequence-to-sequence models, which enable us to leverage powerful and biomedical domain pre-trained language models and apply different strategies to generate lay summaries from long documents. We conducted various experiments to carefully investigate the effects of different aspects of this long-document summarization task such as extracting different document lengths and utilizing different pre-trained language models. We achieve the third rank in the shared task (and the second rank excluding the baseline submission of the organizers).