Invited by the Key Laboratory of Speech Acoustics and Content Understanding of Institute of Acoustics (IOA), ZHANG Shi-Xiong from the University of Cambridge visited IOA and shared his research interest of structured support vector machines (S-SVM) for continuous speech recognition on June 21.

ZHANG gave a lecture during the academic exchanges. The lecture described an S-SVM approach in a specialized framework suitable for medium to large vocabulary speech recognition. Besides, an important aspect of S-SVMs is the form of the joint feature spaces. Here, generative models, hidden Markov models, are used to obtain the features. By using generative models to derive the features, state-of-the-art model-based compensation schemes can be used to make the system robust to noise. Afterwards, the performance of S-SVMs is evaluated on noise robust speech recognition tasks: AURORA 2and 4.

The theory and features of S-SVM for continuous speech recognition were heatedly discussed by the attendees from IOA.

ZHANG Shi-Xiong’s research interests include speech recognition, speaker verification and machine learning. He was nominated a 2011 Interspeech Best Student Paper Award for his paper “Structured Support Vector Machines for Noise Robust Continuous Speech Recognition”. In addition, he was awarded Best Paper Award in 2008 IEEE Signal Processing Postgraduate Forum for his paper "Articulatory-Feature based Sequence Kernel for High-Level Speaker Verification”.

ZHANG Shi-Xiong from the University of Cambridge