Comparative Study of Linear Feature Transformation Techniques for Mandarin Digit String Recognition
نویسندگان
چکیده
Linear feature transformation technique is widely used to improve feature discriminability. It can reduce the dimensionality of the feature space, un-correlate the feature components, hence more discriminative model can be obtained. In this paper we compare three discriminative linear transformation approaches in Mandarin digit string recognition (MDSR) system. Compared with the conventional Linear Discriminant Analysis (LDA), two other discriminative linear transformation methods derived from LDA, that is Confusion Discriminant Analysis (CDA) and Heteroscedastic Discriminant Analysis (HDA), are studied on the basis of state-specific confusable class definition and its class-dependent linear transformations.
منابع مشابه
Improve the Implementation of Pitch Features for Mandarin Digit String Recognition Task
Mandarin digit string recognition (MDSR) is a difficult task in the field of automatic speech recognition (ASR) and using pitch feature can significantly improve the performance. In conventional methods of pitch feature extraction, random value is commonly used as pitch output in unvoiced (UV) frames, which causes serious statistical confusion between voiced (V) and UV units and incurs abnormal...
متن کاملAn Efficient Method for Removing Deletion Errors in Quickly-spoken Connected Mandarin Digit String Speech Recognition
Connected Mandarin digit string speech, especially at rapid spoken rate, is very difficult to recognize correctly. In this paper, a new training method named neighboring digits pattern is proposed in order to eliminate most of deletion errors which frequently occur in Mandarin digits speech recognition at high speaking rate when we have enough quickly-spoken speech data as the training set. The...
متن کاملA Comparative Study on Wavelet Packet Based Front-end in Connected Mandarin Digit Recognition
This paper investigates the wavelet packet based front-ends for the connected mandarin digit recognition task. Firstly an ERBlike wavelet packet basis is proposed. Then two kinds of wavelets are selected for comparison. One is the Vaidyanathan wavelet, which has good frequency selectivity but big shift variance. The other is the reverse biorthogonal spline wavelet with excellent shift invariant...
متن کاملDuration Modeling in Mandarin Connected Digit Recognition
Digit string recognition is required in many applications which need to recognize numbers such as telephone numbers, credit card numbers, date, etc. In order to design a high performance recognizer, duration information is explored in this study. In a Mandarin connected digit recognizer, insertion and deletion errors amount to more than two thirds of the total recognition errors because there e...
متن کاملNoise Suppression Based on Teager Energy Operator for Improving the Robustness of Asr Front-end
In this paper, we proposed a new noise suppression method based on Teager Energy Operator in advancing the noise robustness of speech recognition front-end. The presented method attempts to remove a distortion estimation in Teager energy domain, especially, a Teager energy estimation of noise signal is subtracted from the noisy speech signal. This approach differs significantly from the traditi...
متن کامل