The One About Embeddings & Speech AI

How do AI models represent what we mean, and understand the way we speak? Join researchers and practitioners as they explore embedding and speech models, from research, evaluation, and real-world applications to adapting models for multilingual context. Hear about the approaches and trade-offs behind building models that work better for the languages, speech, and contexts they're used in. More info to come soon, keep a lookout! 👀 More About the Sharings Raymond & Peerat (AI Singapore) will share on “Building and Evaluating Embedding Models for Southeast Asia.” How well do today's embedding models actually work across Southeast Asian languages? Raymond and Peerat will share their work on SEA-BED, a benchmark for evaluating embedding models across Asian languages, alongside insights from the pre-training and post-training of SEA-LION-ModernBERT and SEA-LION Embedding models. They'll then be joined by Vincent (Lead Data Scientist, GoTo Group), an industry partner on the work, who'll bring in the production perspective and share practical considerations and lessons from deploying embedding models in real-world applications. (Technical Level: 200) Dr Ivan Kukanow & Zheng Xin (KLASS Engineering & Solutions) will share on "Singlish, Can or Not? Fine-Tuning and Evaluating Zero-Shot TTS for Singapore English" (“Speech systems by KLASS”) As text-to-speech systems become increasingly natural, getting them to faithfully represent different accents and linguistic identities remains a challenge. Ivan and Zheng Xin will share their work on adapting zero-shot text-to-speech (ZS-TTS) models to better capture Singapore English, moving beyond generic English pronunciation towards more locally representative speech. Drawing on their work accepted at SLT 2026, they'll explore the approaches and challenges behind preserving speaker identity and regional accents in speech synthesis, and what they've learnt about adapting these systems for Singapore's multilingual and multicultural context. They'll also discuss how these approaches could extend to other underrepresented accents and language communities. (Technical Level: 200) More About the Speakers Raymond Ng is an AI engineer with the SEA-LION team in AI Products, AI Singapore. Raymond mainly contributes to pre-training and continued pre-training effort for the SEA-LION project (LLM and embedding models) as well as research work on Multilingual Large Language Models. Raymond has a Master of Technology (MTech), Intelligent System from National University of Singapore (NUS). Peerat Limkonchotiwat is a Research Fellow at AI Product, AI Singapore, under the host of NTU. His responsibility is to develop and research around Southeast Asian (SEA) NLP, such as SEA-LION, SEA-HELM, SEA-Guard, and SEA-LION Embedding. Those models are industry-centric and reflect the needs in SEA, including advances in LLMs, safety, and embeddings for RAG. Moreover, he also works as the main contributor of SEACrowd, a non-profit organization project that collects data in SEA and does SEA research, such as SEACrowd (EMNLP’24) and SEA-VL (ACL’25). In addition, he is also an Affiliated Researcher at Chulalongkorn University, Thailand, and a member of the advisory board of SIGSEA. Vincent Nguyen is a Lead Data Scientist at GoTo Group in Singapore, where he leads GenAI and multimodal LLM initiatives. Previously, he was a Postdoctoral AI Researcher at CSIRO’s Data61 in Australia, applying NLP to climate adaptation and healthcare. He holds a Ph.D. in Computer Science from the Australian National University (ANU) and a B.S. (Adv) (Hons) from the University of Sydney. His work focuses on bringing the latest academic research in foundation models, representation learning, and production AI to large-scale, production-grade use cases with regional impact. Dr Ivan Kukanov is a Principal Researcher at KLASS Engineering and Solutions, Singapore. He holds a PhD from the University of Eastern Finland in Science, Technology and Computing, currently leading research in audio-visual deepfake and spoofing detection. Formerly of the Institute for Infocomm Research (I²R), A*STAR Singapore, his research work also includes emotion recognition, speaker verification, speech translation. He has published in top venues such as ICASSP, INTERSPEECH, and IEEE TASLP, as well as international evaluations such as the 1M-Deepfake Detection Challenge in 2025, placing 3rd internationally. Dr Kukanov has also contributed to the research community as a co-organizer of the ASVspoof 5 and RADAR-2026 Challenges, contributing to the international benchmarking of spoofing and deepfake detection systems. Zheng Xin Chai is a Machine Learning Engineer at KLASS Engineering and Solutions, Singapore, with a decade of experience spanning applied research and engineering in the defence science and homeland security domains. His work focuses primarily on speech and natural language processing, developing and translating machine learning technologies into practical applications. His recent work includes automatic speech recognition, audio-visual speech recognition (AVSR), and text-to-speech (TTS), including adapting speech technologies to Singaporean English. His past work explored web-adaptive information extraction, visual question answering (VQA), and multimodal AI systems. He has contributed to the development and evaluation of end-to-end machine learning pipelines, including assessing open-source solutions against defined technical requirements and integrating them into practical applications. More About The Series AI Wednesdays is Lorong AI’s weekly gathering, bringing together practitioners, researchers and innovators for technical discussions on research insights, product development and engineering practices. Get involved: Learn more about Lorong AI | Speaker Sign-up | WhatsApp Community | LinkedIn | X