30/11/2025
This week we had the pleasure of welcoming Dr. Tanel Alumäe for a fascinating guest lecture on speech recognition! 🙌
He opened by reminding us that while speech recognition has come a long way, it’s not the same as speech understanding. These systems can convert speech to text with impressive accuracy, but they still don’t truly grasp the meaning of a conversation. 👏 Their strengths lie in tasks like dictation, transcribing meetings or lectures, and processing call audio—places where turning spoken words into text is the goal.🧐
Dr. Alumäe also highlighted why this field is so challenging: human speech is incredibly variable. We rarely pronounce words as neatly as they appear in writing, and countless factors influence how we speak—our physiology, speaking style, accents, and even our surroundings. For example, the Lombard effect causes us to instinctively speak more clearly when we’re in a noisy environment. 🗣️
To handle this complexity, modern speech-processing pipelines rely on tools like the Discrete Fourier Transform, short-time spectral analysis, and spectrograms—techniques that let models capture the key features hidden inside the audio signal. Professor Alumäe also walked us through today’s dominant end-to-end speech recognition models and how they build directly from these features to final text output.
A big thank-you to everyone who joined us, and especially to Tanel Alumäe for an insightful and engaging lecture! 😌
🧑🏫Tanel Alumüe is a tenured associate professor and the head of the Laboratorz of Language Technologz at Tallinn University of Technology (TalTech - Tallinna Tehnikaülikool). Professor Alumüe's specialization is related to speech technology. 👨💻
📌This guest lecture was organised as part of the study course “Applications of Language Technologies" on November 27, 2025 at 4:30 PM in the 16th auditorium of the main building of the University of Latvia (Raiņa bulvāris 19).
👉The guest lecture was implemented with the financial support of the "Language Technology Initiative". The project "Language Technology Initiative" (No. 2.3.1.1.i.0/1/22/I/CFLA/002) is co-financed by the European Union's Recovery and Resilience Facility Investment and the State Budget.
