Hough transform based masking in feature extraction for noisy speech recognition

Despite various advances in recent years, robustness in the presence of various types and levels of environmental noise remains a critical issue for automatic speech recognition systems. This paper describes a novel and noise robust front-end that incorporates the use of Hough transform for simultaneous frequency and temporal masking, together with cumulative distribution mapping of cepstral coefficients, for noisy speech recognition. Recognition experiments on the Aurora II connected digits database have revealed that the proposed front-end achieves an average digit recognition accuracy of 83.31% for all the three Aurora test sets. Compared with the recognition results obtained by using the ETSI standard Mel-cepstral front-end, this accuracy represents a relative error rate reduction of around 57%.