• Worry not about background issues. Fundamental courses are offered.
• Strong motivation and work ethics are indeed required.

Automatic Speech Recognition

From input speech signal to output text, automatic speech recognition (ASR) is one of the most important and most difficult problem for computer scientists and engineers to solve. Imagine your computer would just listen to you! How convenient would that be?

OK, then why are voice-activating computers not everywhere? ASR systems suffer from the presence of noises. This is exactly the focus of our researches: the noise-robustness of ASR systems.

In addition to ASR, we also have students working on speaker diarization, language identification and speech synthesis. They are all parts of a general field called audio, speech and language processing. We welcome students who are interested in this field.

Speech technology is generally considered as the mainstream of next generation personal computers and mobile phones. In Taiwan, big-name companies such as Acer, Chung-hua Telecom, to name a few, have research teams in speech technology. "My ET" is based on speech technology as well. You may not know that Cyberon, which is based on Taiwan, is one of the top cellphone ASR engine providers in the world.

Machine Translation

From an input sentence in one language to an output sentence in another language, machine translation (MT) is a dream application which tunnels through language barriers. For example, MT from English to Chinese enables people who read Chinese to acquire information written in English, and vice versa. One can think of many scenarios in which such systems can be quite useful.

Believe it or not, many people working on MT come from ASR community. This is because the very similar fundamental principles they use in solving their respective problems. In the statistical framework, MT and ASR both involve modeling, parameter estimation, and optimal hypothesis searching problems.

Comparing to ASR, MT appears to be more versatile in its statistical models. In a way, this also makes the subject more challenging. We welcome courageous students to join us in this adventure.

Machine Learning

Classification, recognition, and clustering are typical problems of an area called machine learning. Whenever a problem is to be solved automatically by computers, we have a machine learning problem. Indeed, many real-world problems can be abstracted as machine learning problems.

Features, models, hypothesis searching algorithms are the cornerstones for machine learning. We are interested in machine learning because it is general and useful. In fact, ASR and MT can be seen as problems in machine learning. Suppose you have a problem of a classification or detection nature to solve, chances are that you will have better ideas of where to start if you become acquainted with certain machine learning approaches.