Written by Vadrianey Anak Asas & Ts Dr Sarah Flora Samson Juan
For the past three decades, the Center for Language and Speech Processing (CLSP) at Johns Hopkins University has been organizing an annual six-week international research workshop on speech and language engineering, with a strong emphasis in recent years on machine learning (ML) and artificial intelligence (AI).
These workshops had made a significant contribution to the Human Language Technology (HLT) community, influencing both academic research and industrial applications.
The 32nd edition of the workshop, JSALT 2025, was held in Brno, Czechia, from June 23 to August 1, 2025. It was co-hosted by the Speech@FIT group at the Faculty of Information Technology, Brno University of Technology, together with Phonexia.
During the program, participants worked in dedicated teams that engaged intensively in addressing challenging research problems or exploring novel techniques that had not yet been fully investigated.
The workshops fostered a collaborative environment where participants formed long-term international research partnerships.
The workshops fostered a highly collaborative environment where participants benefited from close interactions with international peers, often leading to long-term research partnerships and collaborations.
In 2025, the workshop comprised four research teams, each focusing on a distinct area of speech and language technology:
• Play Your Part: Towards LLM role-playing agents that stick to their role
• Advancing Expert-Level Reasoning and Understanding in Large Audio Language Models
• EMMA: End-to-End Multi-Channel Multi-Talker ASR
• TTS4ALL: TTS in low-resource scenarios, covering data management, methodology, models, and evaluation
Ts. Dr Sarah Flora, a senior lecturer and Vadrianey Asas, a postgraduate student from Faculty of Computer Science and Information Technology (FCSIT), UNIMAS were invited to participate in the workshop.
They arrived on June 22 for the six-week intensive research program, which brought together international teams to tackle emerging challenges in speech and language technologies.
They were part of the TTS4ALL team, officially titled “TTS in Low-Resource Scenarios: Data Management, Methodology, Models, and Evaluation”, led by Dr. Yannick Estève and Dr. Fethi Bougares.
They were heavily involved in data qualification and evaluation, as well as investigating effective strategies for training text-to-speech (TTS) systems using low-resource language data.
In addition to intensive project work, the workshop featured a series of weekly plenary lectures delivered by leading experts in the field, exposing participants to the latest advancements in machine learning, speech processing, and natural language technologies.
Each team also conducted daily progress meetings and weekly updates, ensuring collaboration and alignment across projects.
Beyond research activities, the workshop provided a platform for networking and cultural exchange through a variety of social events, enabling researchers from diverse backgrounds to share experiences and foster long-term collaborations.
The program culminated in a final presentation week, during which each team showcased their research findings and outcomes to the broader community.
The participation of UNIMAS in JSALT 2025 not only highlighted the university’s active involvement in cutting-edge research on speech and language technologies but also strengthened its international presence in the field of low-resource language processing.
This engagement is expected to pave the way for future collaborations with leading global institutions and contribute to advancing research capacity in Malaysia.
Researchers participating in JSALT 2025
Members of TTS4ALL team
FCSIT is one of the Premier Digital Tech Institutions endorsed by Malaysia Digital Economy Corporation (MDEC).
#mypdti