Sponsored by the IEEE Signal Processing Society
For contributions to speech processing and feature extraction for robust speech recognition
The pioneering speech processing methods developed by Hynek Hermansky are some of the most widely used in speech recognition, speech and audio coding, and speaker and language recognition. Known for innovative signal processing approaches that make speech parameters insensitive to channel variability and noise, his many accomplishments include Perceptual Linear Prediction (PLP) signal representation, Relative Spectral Analysis (RASTA) processing, and TRAP and TANDEM neural net based feature extraction techniques, in machine recognition of speech. PLP is one of the two primary feature sets used internationally for speech recognition. RASTA helps to reduce the effects of linear spectral distortion when using a microphone for speech recognition with a different frequency response than what was used during training. TRAP is one of the first techniques which explicitly uses very large temporal contexts of speech and TANDEM allows for interface between modern neural net features and conventional speech recognition systems.
An IEEE Fellow, Hermansky is the Julian S. Smith Professor of Electrical Engineering and director of the Center for Language and Speech Processing at Johns Hopkins University, Baltimore, MD, USA.