The international research workshop JSALT 2025 has come to a close

The JSALT 2025 international research workshop, organized by the Center for Language and Speech Processing at Johns Hopkins University in Maryland, concluded with a gala reception at the BUT Rector's Office on Friday, August 1. The event, which this year attracted more than 100 top researchers in the field of speech and language technologies from 22 countries, took place at the Faculty of Information Technology BUT over a period of six weeks (preceded by a summer school). The core of the program was the scientific work of four research teams, which presented their findings in the final days of the workshop. A unifying theme across the teams was machine learning applied to human speech. The first team focused on improving the ability of large language models to consistently adhere to assigned roles (doctor, counselor, etc.), demonstrating, for example, the importance of persona-based prompting and developing a framework for sound synthesis to generate spoken dialogue using LLM. The outputs of the second team, which focused on language models working with sound, included new audio datasets covering a range of areas to support preliminary model training and a new methodology to assess answer correctness. The third team, working on simplifying and streamlining speech recognition in complex scenarios, presented e.g. an improved speaker diarization conditioned speech transcription. The last team, for example, came up with a new toolkit for evaluating text-to-speech (TTS) conversion for low-resource languages (e.g., Bengali, Kurdish, or Sarawak Malay) and offered new training and testing data for these types of languages.

Although we are a small country, speech and language technologies rank us among the most important global centers, alongside the United States and China. The Czech Republic, and specifically Brno, has made an indisputable mark in this field of artificial intelligence thanks to the Speech@FIT research group and names such as Hynek Heřmanský, Jan Černocký, and Lukáš Burget. This is one of the reasons why we were able to welcome personalities such as Ricard Marxer (University of Toulon), Ramani Duraiswami (University of Maryland), Samuele Cornell (Carnegie Mellon University), Yannick Estève (Laboratoire d'Informatique d'Avignon), Fethi Bougares (Elyadata), and the main organizer of the event, Sanjeev Khudanpur (Johns Hopkins University). In fact, the workshop offered seven plenary lectures by leading experts, which were also aimed at a wider audience.

The aim of this prestigious event is, of course, to push the boundaries of research. However, perhaps an even more significant outcome of JSALT can be seen in its support for international scientific cooperation and the establishment of new interpersonal contacts (and thus contacts between institutions). Many of today's technologies have their roots in research and collaborations that began at JSALT. One example is the automatic translation system found in Google Translate. „Brno is a recognized center of European research in the field of speech and language artificial intelligence, but from mid-June to early August, this position was tangible, visible, and audible – the concentration of so many smart colleagues, talented doctoral and master's students, invited speakers, important visitors from universities and industry, presentations, brainstorming sessions, daily meetings, and the exchange of ideas made JSALT a very intense experience. We are very pleased that our participants took home not only technical know-how, contacts with colleagues, a foundation for further scientific work, articles, software, and data, but also that they became attached to FIT VUT and Brno as suitable places for work and entertainment. Brno is magnetic in the best sense of the word, and I think we will see many of them here again,“ says Jan „Honza“ Černocký, evaluating this year's JSALT.

All of the above points can be considered fulfilled this year, with a significant contribution from FIT BUT and its "speech data miners". We would like to thank the organizers, presenters, and participants for their work and enthusiasm!

You can find a photo gallery from the final days of the workshop HERE (photos: Martin Horný).