Search

This paper presents the deep neural networks to classification of children with voice impairments from speech signals. In the analysis of speech signals, 6,373 static acoustic features are extracted from many kinds of low-level-descriptors and functionals. To reduce the variability of extracted features, two-dimensional normalizations are applied to smooth the interspeaker and inter-feature mismatch using the feature warping approach. Then, the feature selection is used to explore the discriminative and low-dimensional representation based on techniques of principal component analysis and linear discriminant analysis. In such representation, the robust features are obtained by eliminating noise features via subspace projection. Finally, the deep neural networks are adopted to classify the children with voice impairments. We conclude that deep neural networks with the proposed feature normalization and selection can significantly contribute to the robustness of recognition in practical application scenarios. We have achieved an UAR of 60.9% for the four-way diagnosis classification on the development set. This is a relative improvement of 16.2% to the official baseline by using our single system. [ABSTRACT FROM PUBLISHER]

This study presents a noise robust front-end postprocessing technology. After cepstral feature analysis, the feature normalization is usually applied for noisy reduction in spoken language recognition. We investigate a highly effective MVAW processing based on standard MFCC and SDC features on NIST-LRE 2007 tasks. The procedure includes mean subtraction, variance normalization, auto-regression moving-average filtering and feature warping. Experiments were conducted on a common GMM-UBM system. The results indicated significant improvements in recognition accuracy. [ABSTRACT FROM PUBLISHER]

Use of a linear projection (LP) function to transform multiple sets of acoustic models into a single set of acoustic models is proposed for characterizing testing environments for robust automatic speech recognition. The LP function is an extension of the linear regression (LR) function used in maximum likelihood linear regression (MLLR) and maximum a posteriori linear regression (MAPLR) by incorporating local information in the ensemble acoustic space to enhance the environment modeling capacity. To estimate the nuisance parameters of the LP function, we developed maximum likelihood LP (MLLP) and maximum a posteriori LP (MAPLP) and derived a set of integrated prior (IP) densities for MAPLP. The IP densities integrate multiple knowledge sources from the training set, previously seen speech data, current utterance, and a prepared tree structure. We evaluated the proposed MLLP and MAPLP on the Aurora-2 database in an unsupervised model adaptation manner. Experimental results show that the LP function outperforms the LR function with both ML- and MAP-based estimates over different test conditions. Moreover, because the MAP-based estimate can handle over-fittings well, MAPLP has clear improvements over MLLP. Compared to the baseline result, MAPLP provides a significant 10.99% word error rate reduction. [ABSTRACT FROM PUBLISHER]

In this paper we perform a comparison of lookahead composition and on-the-fly hypothesis rescoring using a common decoder. The results on a large vocabulary speech recognition task illustrate the differences in the behaviour of these algorithms in terms of error rate, real time factor, memory usage and internal statistics of the decoder. The evaluations were performed when the decoder was operated at either the state or arc level. The results show the dynamic approaches also work well at the state level even though there is greater dynamic construction cost. [ABSTRACT FROM PUBLISHER]

In this paper, we present our work on collecting spontaneous texts from the Web for constructing a language model in a Chinese speech recognition system. The selection of spontaneous-like texts involves two steps: First, word-segmented web texts are selected using a perplexity-based approach in which the style-related words are strengthened by omitting infrequent topic words from similarity measurements. Second, the selected texts are then clustered based on non-noun part-of-speech (POS) words and optimal clusters are chosen by referring to a set of spontaneous seed sentences. Using the language model interpolated with the one trained by the selected sentences and a baseline model, speech recognition evaluations were conducted on an open domain spontaneous test set. We effectively reduced the character error rate (CER), with 1.64% absolute (or 6.5% relative) reduction by comparison with the baseline model. We also verified that the proposed method is superior to the conventional perplexity-based approach with about 1% absolute (or 4.0% relative) reduction in CER. [ABSTRACT FROM PUBLISHER]

Ensemble acoustic modeling can be used to model different factors that cause variability of acoustic space, and provide different combination to improve the performance of automatic speech recognition (ASR). One of the main concerns is how to partition the training data set to several subsets based on which ensemble models are trained. In this study, we focus on ensemble acoustic modeling concerned with acoustic variability caused by gender and accent for Chinese large vocabulary continuous speech recognition (LVCSR). Considering that gender and accent information may be encoded in local acoustic realizations of a few specific phonetic classes rather than in a global acoustic distribution, we proposed a acoustic space partition method based on broad phonetic class (BPC) modeling of speaker for ensemble acoustic modeling. With the principal component analysis (PCA) of the BPC based speaker representation, we designed two level hierarchical data partitions in the low dimensional speaker factor space that concerned with gender and accent information. Ensemble acoustic models were trained on the partitioned data sets on both levels. Speech recognition results showed that using acoustic models trained based on the first level and second level partitions got 9.73% and 32.29% relative improvements in character error reduction rate, respectively. [ABSTRACT FROM PUBLISHER]

The tradeoff between noise reduction and speech distortion is a key concern in designing noise reduction algorithms. We have proposed a regularization framework for noise reduction with the consideration of the tradeoff problem. We regard speech estimation as a functional approximation problem in a reproducing kernel Hilbert space (RKHS). In the estimation, the objective function is formulated to find an approximation function that gives a good tradeoff between the approximation accuracy and complexity of the function. By using a regularization method, the approximation function can be estimated from noisy observations. In this paper, we further provided a theoretical analysis of the tradeoff property of the framework in noise reduction. We applied the framework for speech enhancement experiments in real applications. Compared with several classical noise reduction methods, the proposed framework showed promising advantages. [ABSTRACT FROM PUBLISHER]