Multi-task learning in deep neural networks has been proven to be effective for acoustic modeling in speech recognition. In the paper, this technique is applied to Mandarin-English code-mixing recognition. For the primary task of the senone classification, three schemes of the auxiliary tasks are proposed to introduce the language information to networks and improve the prediction of language switching. On the real-world Mandarin-English test corpus in mobile voice search, the proposed schemes enhanced the recognition on both languages and reduced the relative overall error rates by 3.5%, 3.8% and 5.8% respectively.

In this paper, a logarithmic adaptive quantization projection (LAQP) algorithm for digital watermarking is proposed. Conventional quantization index modulation uses a fixed quantization step in the watermarking embedding procedure, which leads to poor fidelity. Moreover, the conventional methods are sensitive to value-metric scaling attack. The LAQP method combines the quantization projection scheme with a perceptual model. In comparison to some conventional quantization methods with a perceptual model, the LAQP only needs to calculate the perceptual model in the embedding procedure, avoiding the decoding errors introduced by the difference of the perceptual model used in the embedding and decoding procedure. Experimental results show that the proposed watermarking scheme keeps a better fidelity and is robust against the common signal processing attack. More importantly, the proposed scheme is invariant to value-metric scaling attack.

Traditional two-microphone noise reduction algorithms to deal with highly nonstationary directional noises generally use the direction of arrival or phase difference information. The performance of these algorithms deteriorate when diffuse noises coexist with nonstationary directional noises in realistic adverse environments. In this paper, we present a two-channel noise reduction algorithm using a spatial information-based speech estimator and a spatial-information-controlled soft-decision noise estimator to improve the noise reduction performance in realistic non-stationary noisy environments. A target presence probability estimator based on Bayes rules using both phase difference and magnitude squared coherence is proposed for soft-decision of noise estimation, so that they can share complementary advantages when both directional noises and diffuse noises are present. Performances of the proposed two-microphone noise reduction algorithm are evaluated by noise reduction, log-spectral distance (LSD) and word recognition rate (WRR) of a distant-talking ASR system in a real room's noisy environment. Experimental results show that the proposed algorithm achieves better noises suppression without further distorting the desired signal components over the comparative dual-channel noise reduction algorithms.

This paper presents our investigation into improving the performance of our previous automatic reading quality assessment system. The method of the baseline system is calculating the average value of the Phone Log-Posterior Probability (PLPP) of all phones in the voice to be assessed, and the average value is used as the reading quality assessment feature. In this paper, we presents three improvements. First, we cluster the triphones, and then calculate the average value of the normalized PLPP for each classification separately, and use this average values as the multi-dimensional assessment features instead of the original one-dimensional assessment feature. This method is simple but effective, which made the score difference of the machine scoring and manual scoring decrease by 30.2% relatively. Second, in order to assess the reading rhythm, we train Gaussian Mixture Models (GMM), which contain the information of each triphone's relative duration under standard pronunciation. Using the GMM, we can calculate the probability that the relative duration of each phone is conform to the standard pronunciation, and the average value of the probabilities is added to the assessment feature vector as a dimension of feature, which decreased the score difference between the machine scoring and manual scoring by 9.7% relatively. Third, we detect Filled Pauses (FP) by analyzing the formant curve, and then calculate the relative duration of FP, and add the relative duration of FP to the assessment feature vector as a dimension of feature. This method made the score difference between the machine scoring and manual scoring be further decreased by 10.2% relatively. Finally, when the feature vector extracted by the three methods are used together, the score difference between the machine scoring and manual scoring was decreased by 43.9% relatively compared to the baseline system.

This paper describes a reading miscue detection system based on the conventional Large Vocabulary Continuous Speech Recognition (LVCSR) framework [1]. In order to incorporate the knowledge of reference (what the reader ought to read) and some error patterns into the decoding process, two methods are proposed: Dynamic Multiple Pronunciation Incorporation (DMPI) and Dynamic Interpolation of Language Model (DILM). DMPI dynamically adds some pronunciation variations into the search space to predict reading substitutions and insertions. To resolve the conflict between the coverage of error predications and the perplexity of the search space, only the pronunciation variants related to the reference are added. DILM dynamically interpolates the general language model based on the analysis of the reference and so keeps the active paths of decoding relatively near the reference. It makes the recognition more accurate, which further improves the detection performance. At the final stage of detection, an improved dynamic program (DP) is used to align the confusion network (CN) from speech recognition and the reference to generate the detecting result. The experimental results show that the proposed two methods can decrease the Equal Error Rate (EER) by 14% relatively, from 46.4% to 39.8%.

This paper focuses on the problem of performance degradation in mismatched speech recognition. The F-Ratio analysis method is utilized to analyze the significance of different frequency bands for speech unit classification, and we find that frequencies around 1 kHz and 3 kHz, which are the upper bounds of the first and the second formants for most of the vowels, should be emphasized in comparison to the Mel-frequency cepstral coefficients (MFCC). The analysis result is further observed to be stable in several typical mismatched situations. Similar to the Mel-Frequency scale, another frequency scale called the F-Ratio-scale is thus proposed to optimize the filter bank design for the MFCC features, and make each subband contains equal significance for speech unit classification. Under comparable conditions, with the modified features we get a relative 43.20% decrease compared with the MFCC in sentence error rate for the emotion affected speech recognition, 35.54%, 23.03% for the noisy speech recognition at 15 dB and 0 dB SNR (signal to noise ratio) respectively, and 64.50% for the three years' 863 test data. The application of the F-Ratio analysis on the clean training set of the Aurora2 database demonstrates its robustness over languages, texts and sampling rates.

This letter evaluates the relative contributions of temporal fine structure cues in various frequency bands to Mandarin tone perception using novel "auditory chimaeras". Our results confirm the importance of temporal fine structure cues to lexical tone perception and the dominant region of lexical tone perception is found, namely the second to fifth harmonics can contribute no less than the fundamental frequency itself.

The GMM-UBM framework has been proved to be one of the most effective approaches to the automatic speaker verification (ASV) task in recent years. In this letter, we first propose an approximate decision function of traditional GMM-UBM, from which it is shown that the contribution to classification of each Gaussian component is equally important. However, research in speaker perception shows that a different speech sound unit defined by Gaussian component makes a different contribution to speaker verification. This motivates us to emphasize some sound units which have discriminability between speakers while de-emphasize the speech sound units which contain little information for speaker verification. Experiments on 2006 NIST SRE core task show that the proposed approach outperforms traditional GMM-UBM approach in classification accuracy.

The simultaneous utterances impact the ability of both the hearing-impaired persons and automatic speech recognition systems. Recently, deep neural networks have dramatically improved the speech separation performance. However, most previous works only estimate the speech magnitude and use the mixture phase for speech reconstruction. The use of the mixture phase has become a critical limitation for separation performance. This study proposes a two-stage phase-aware approach for multi-talker speech separation, which integrally recovers the magnitude as well as the phase. For the phase recovery, Multiple Input Spectrogram Inversion (MISI) algorithm is utilized due to its effectiveness and simplicity. The study implements the MISI algorithm based on the mask and gives that the ideal amplitude mask (IAM) is the optimal mask for the mask-based MISI phase recovery, which brings less phase distortion. To compensate for the error of phase recovery and minimize the signal distortion, an advanced mask is proposed for the magnitude estimation. The IAM and the proposed mask are estimated at different stages to recover the phase and the magnitude, respectively. Two frameworks of neural network are evaluated for the magnitude estimation on the second stage, demonstrating the effectiveness and flexibility of the proposed approach. The experimental results demonstrate that the proposed approach significantly minimizes the distortions of the separated speech.

In this paper, we present a hybrid speech emotion recognition system exploiting both spectral and prosodic features in speech. For capturing the emotional information in the spectral domain, we propose a new spectral feature extraction method by applying a novel non-uniform subband processing, instead of the mel-frequency subbands used in Mel-Frequency Cepstral Coefficients (MFCC). For prosodic features, a set of features that are closely correlated with speech emotional states are selected. In the proposed hybrid emotion recognition system, due to the inherently different characteristics of these two kinds of features (e.g., data size), the newly extracted spectral features are modeled by Gaussian Mixture Model (GMM) and the selected prosodic features are modeled by Support Vector Machine (SVM). The final result of the proposed emotion recognition system is obtained by combining the results from these two subsystems. Experimental results show that (1) the proposed non-uniform spectral features are more effective than the traditional MFCC features for emotion recognition; (2) the proposed hybrid emotion recognition system using both spectral and prosodic features yields the relative recognition error reduction rate of 17.0% over the traditional recognition systems using only the spectral features, and 62.3% over those using only the prosodic features.

Morphemes, which are obtained from morphological parsing, and statistical sub-words, which are derived from data-driven splitting, are commonly used as the recognition units for speech recognition of agglutinative languages. In this letter, we propose a discriminative approach to select the splitting result, which is more likely to improve the recognizer's performance, for each distinct word type. An objective function which involves the unigram language model (LM) probability and the count of misrecognized phones on the acoustic training data is defined and minimized. After determining the splitting result for each word in the text corpus, we select the frequent units to build a hybrid vocabulary including morphemes and statistical sub-words. Compared to a statistical sub-word based system, the hybrid system achieves 0.8% letter error rates (LERs) reduction on the test set.

In recent decades, there has been a great deal of research into the problem of bilingual speech recognition - to develop a recognizer that can handle inter- and intra-sentential language switching between two languages. This paper presents our recent work on the development of a grammar-constrained, Mandarin-English bilingual Speech Recognition System (MESRS) for real world music retrieval. Two of the main difficult issues in handling the bilingual speech recognition systems for real world applications are tackled in this paper. One is to balance the performance and the complexity of the bilingual speech recognition system; the other is to effectively deal with the matrix language accents in embedded language. In order to process the intra-sentential language switching and reduce the amount of data required to robustly estimate statistical models, a compact single set of bilingual acoustic models derived by phone set merging and clustering is developed instead of using two separate monolingual models for each language. In our study, a novel Two-pass phone clustering method based on Confusion Matrix (TCM) is presented and compared with the log-likelihood measure method. Experiments testify that TCM can achieve better performance. Since potential system users' native language is Mandarin which is regarded as a matrix language in our application, their pronunciations of English as the embedded language usually contain Mandarin accents. In order to deal with the matrix language accents in embedded language, different non-native adaptation approaches are investigated. Experiments show that model retraining method outperforms the other common adaptation methods such as Maximum A Posteriori (MAP). With the effective incorporation of approaches on phone clustering and non-native adaptation, the Phrase Error Rate (PER) of MESRS for English utterances was reduced by 24.47% relatively compared to the baseline monolingual English system while the PER on Mandarin utterances was comparable to that of the baseline monolingual Mandarin system. The performance for bilingual utterances achieved 22.37% relative PER reduction.

In this letter, we propose a new approach to SVM based speaker recognition, which utilizes a kind of novel phonotactic information as the feature for SVM modeling. Gaussian mixture models (GMMs) have been proven extremely successful for text-independent speaker recognition. The GMM universal background model (UBM) is a speaker-independent model, each component of which can be considered as modeling some underlying phonetic sound classes. We assume that the utterances from different speakers should get different average posterior probabilities on the same Gaussian component of the UBM, and the supervector composed of the average posterior probabilities on all components of the UBM for each utterance should be discriminative. We use these supervectors as the features for SVM based speaker recognition. Experiment results on a NIST SRE 2006 task show that the proposed approach demonstrates comparable performance with the commonly used systems. Fusion results are also presented.

In this letter, we present an automatic approach of objective singing performance evaluation for untrained singers by relating acoustic measurements to perceptual ratings of singing voice quality. Several acoustic parameters and their combination features are investigated to find objective correspondences of the perceptual evaluation criteria. Experimental results show relative strong correlation between perceptual ratings and the combined features and the reliability of the proposed evaluation system is tested to be comparable to human judges.

In this paper, we presented a novel method for automatic pronunciation quality assessment. Unlike the popular “Goodness of Pronunciation” (GOP) method, this method does not map the decoding confidence into pronunciation quality score, but differentiates the different pronunciation quality utterances directly. In this method, the student's utterance need to be decoded for two times. The first-time decoding was for getting the time points of each phone of the utterance by a forced alignment using a conventional trained acoustic model (AM). The second-time decoding was for differentiating the pronunciation quality for each triphone using a specially trained AM, where the triphones in different pronunciation qualities were trained as different units, and the model was trained in discriminative method to ensure the model has the best discrimination among the triphones whose names were same but pronunciation quality scores were different. The decoding network in the second-time decoding included different pronunciation quality triphones, so the phone-level scores can be obtained from the decoding result directly. The phone-level scores were combined into the sentence-level scores using maximum entropy criterion. The experimental results shows that the scoring performance was increased significantly compared to the GOP method, especially in sentence-level.

In this letter, we adopt a new factor analysis of neighborhood-preserving embedding (NPE) for speaker verification. NPE aims at preserving the local neighborhood structure on the data and defines a low-dimensional speaker space called neighborhood-preserving embedding space. We compare the proposed method with the state-of-the-art total variability approach on the telephone-telephone core condition of the NIST 2008 Speaker Recognition Evaluation (SRE) dataset. The experimental results indicate that the proposed NPE method outperforms the total variability approach, providing up to 24% relative improvement.

Sparse representation has been studied within the field of signal processing as a means of providing a compact form of signal representation. This paper introduces a sparse representation based framework named Sparse Probabilistic Linear Discriminant Analysis in speaker recognition. In this latent variable model, probabilistic linear discriminant analysis is modified to obtain an algorithm for learning overcomplete sparse representations by replacing the Gaussian prior on the factors with Laplace prior that encourages sparseness. For a given speaker signal, the dictionary obtained from this model has good representational power while supporting optimal discrimination of the classes. An expectation-maximization algorithm is derived to train the model with a variational approximation to a range of heavy-tailed distributions whose limit is the Laplace. The variational approximation is also used to compute the likelihood ratio score of all trials of speakers. This approach performed well on the core-extended conditions of the NIST 2010 Speaker Recognition Evaluation, and is competitive compared to the Gaussian Probabilistic Linear Discriminant Analysis, in terms of normalized Decision Cost Function and Equal Error Rate.

Mel Frequency Cepstral Coefficients (MFCC) are the most popular acoustic features used in automatic speech recognition (ASR), mainly because the coefficients capture the most useful information of the speech and fit well with the assumptions used in hidden Markov models. As is well known, MFCCs already employ several principles which have known counterparts in the peripheral properties of human hearing: decoupling across frequency, mel-warping of the frequency axis, log-compression of energy, etc. It is natural to introduce more mechanisms in the auditory periphery to improve the noise robustness of MFCC. In this paper, a k-nearest neighbors based frequency masking filter is proposed to reduce the audibility of spectra valleys which are sensitive to noise. Besides, Moore and Glasberg's critical band equivalent rectangular bandwidth (ERB) expression is utilized to determine the filter bandwidth. Furthermore, a new bandpass infinite impulse response (IIR) filter is proposed to imitate the temporal masking phenomenon of the human auditory system. These three auditory perceptual mechanisms are combined with the standard MFCC algorithm in order to investigate their effects on ASR performance, and a revised MFCC extraction scheme is presented. Recognition performances with the standard MFCC, RASTA perceptual linear prediction (RASTA-PLP) and the proposed feature extraction scheme are evaluated on a medium-vocabulary isolated-word recognition task and a more complex large vocabulary continuous speech recognition (LVCSR) task. Experimental results show that consistent robustness against background noise is achieved on these two tasks, and the proposed method outperforms both the standard MFCC and RASTA-PLP.

The optimization of spoken dialog management policies is a non-trivial task due to the erroneous inputs from speech recognition and language understanding modules. The dialog manager needs to ground uncertain semantic information at times to fully understand the need of human users and successfully complete the required dialog tasks. Approaches based on reinforcement learning are currently mainstream in academia and have been proved to be effective, especially when operating in noisy environments. However, in reinforcement learning the dialog strategy is often represented by complex numeric model and thus is incomprehensible to humans. The trained policies are very difficult for dialog system designers to verify or modify, which largely limits the deployment for commercial applications. In this paper we propose a novel framework for optimizing dialog policies specified in human-readable domain language using genetic algorithm. We present learning algorithms using user simulator and real human-machine dialog corpora. Empirical experimental results show that the proposed approach can achieve competitive performance on par with some state-of-the-art reinforcement learning algorithms, while maintaining a comprehensible policy structure.

Introducing pronunciation models into decoding has been proven to be benefit to LVCSR. In this paper, a discriminative pronunciation modeling method is presented, within the framework of the Minimum Phone Error (MPE) training for HMM/GMM. In order to bring the pronunciation models into the MPE training, the auxiliary function is rewritten at word level and decomposes into two parts. One is for co-training the acoustic models, and the other is for discriminatively training the pronunciation models. On Mandarin conversational telephone speech recognition task, compared to the baseline using a canonical lexicon, the discriminative pronunciation models reduced the absolute Character Error Rate (CER) by 0.7% on LDC test set, and with the acoustic model co-training, 0.8% additional CER decrease had been achieved.