Search

This paper presents a front-end enhancement system for automatic speech recognition to address the cocktail party problem. Cocktail party problem is focus on recognizing the target speech when multiple speakers talk in the noisy real-environments. Many conventional techniques have been proposed. In this work, we propose a new framework to integrate the conventional blind source separation and minimum variance distortionless response beamformer for the speech enhancement and source separation of the recent CHiME-5 challenge. In our experiments, we found that the time–frequency (T–F) mask estimation strategy based on the BSS algorithm should be different for speech enhancement and source separation. The main difference is that whether we need to account for background noise as an additional class during T–F mask estimation. Experimental results showed that the proposed framework was very beneficial to improve the speech recognition performance on the Single-array-track of CHiME-5. We obtained relative 13.5% WER reduction than the official baseline system by only improving the front-end speech enhancement framework.

The Transformer has shown impressive performance in automatic speech recognition. It uses the encoder-decoder structure with self-attention to learn the relationship between the high-level representation of the source inputs and embedding of the target outputs. In this paper, we propose a novel decoder structure that features a self-and-mixed attention decoder (SMAD) with a deep acoustic structure (DAS) to improve the acoustic representation of Transformer-based LVCSR. Specifically, we introduce a self-attention mechanism to learn a multi-layer deep acoustic structure for multiple levels of acoustic abstraction. We also design a mixed attention mechanism that learns the alignment between different levels of acoustic abstraction and its corresponding linguistic information simultaneously in a shared embedding space. The ASR experiments on Aishell-1 shown that the proposed structure achieves CERs of 4.8% on the dev set and 5.1% on the test set, which are the best results obtained on this task to the best of our knowledge., Comment: Accepted by INTERSPEECH 2020

In the task of mispronunciation detection, the cross-speaker degradation and some other confusing nuisances are the challenging problems demanding prompt solution. In this paper, we will attempt to remove the non-pronunciation variations in the GLDS-SVM expansion space by using nuisance attribute projection strategy, in order to increase the separating capacity between different phoneme instances. Moreover, different kinds of score normalization methods with softmax, posterior probability vector (PPV), Z-norm and T-norm are comparatively discussed. The experiments on three kinds of speech corpora demonstrate the effectiveness of the above methods, and the performance improvement is not very significant, but sustainable.

In automatic speech grading systems, rare research is followed through addressing the issue of GOR (Goodness Of pRosody). In this paper we propose a novel method by taking the advantage of our QBH (Query By Humming) techniques in 2008 MIREX evaluation task. A set of standard samples related to the top-cream students are initially picked up as templates, a cascade QBH structure is then taken from two metrics: the MOMEL stylization followed by DTW distance; the Fujisaki model followed by EMD distance. Sentence GOR is obtained by the fused confidence between target and each template, and forms a weighted sum as the goodness in the passage level. Experiment results indicate that performance increases with the count of template, and Fujisaki-EMD metric outperforms MOMEL-DTW one in terms of correlation. Their combination can be treated as template based GOR score, compensated with our previous feature based GOR score, the approach can achieve 0.432 in correlation and 17.90% in EER in our corpus. Index Terms: speech prosody, query by humming

Automatic assessment of word stress error is an integral part for oral language grading system. However, problems that the property of vowels depends on its context information and the data sparseness of different vowel class are yet to be solved. This paper shall briefly introduce a hybrid method consisting of both traditional prosodic features and proposed context dependent strategies. In classification word stress is determined by weighting a bottom-up fashioned group tree with modified distributed probability score. In experiment, the overall equal error rate of our proposed system achieves 9.41%, which exhibits relative reduction and its competence of use in stress error detection system.

Mispronunciation detection is an important component in computer assisted language learning (CALL) system. In this work, we introduce an efficient GLDS-SVM based detection method, which is successfully used in language and speaker identification systems, and combine it with traditional methods. The main ideas include: extended MFCC features with normalized formant trajectory information, and then propose a novel multi-model strategy for model training to make full use of samples and solve the problem of data unbalance, finally combine GLDS-SVM method with UBM-GMM system to further improve the performance. Experiments show that GLDS-SVM is highly efficient than traditional RBF-SVM, and the fused system can achieve a significant relative improvement of 17.5% in EER reduction, compared with the baseline UBM-GMM system.

Since it is the most natural way for people to search a specific melody in large music database, query by humming/singing is attracting more and more researcherspsila attention in the field of content-based music information retrieval. In this task, note-based and frame-based similarity measures are two commonly used methods. However, in previous works, researchers always focus on one of the two methods alone. In this paper, we propose a novel scheme taking advantage of two different similarity measurements to improve not only the retrieval accuracy but also the retrieving speed. First, Earth Moverpsilas Distance (EMD), which is note-based and much faster, is adopted to eliminate most unlikely candidate. Then, Dynamic Time Warping (DTW), which is frame-based and more accurate, is executed on these surviving candidates. Finally, fusion strategies of these two similarity measurements are employed to improve the performance of whole system. Experiments show our approach can achieve 92.9% accuracy on the database used in MIREX 2006 QBH contest, which is better than those systems participated in that task.

Although researchers have made great progresses on music genre classification in recent years, the need for more accurate system is still not satisfied. In this paper, we propose a method for further reducing the classification error rate based on multiple classifier fusion. First of all, MFCCs and four features from MPEG-7 audio descriptor are extracted in every short time frame, and then a group of frames are gathered into a longer segment, in which mean and variance of these short time frames features are calculated. The segment is considered as the basic unit for training and testing module. Then random forest (RF) and multilayer perceptron neural network (MLP) are executed on such segment independently. Finally, a weighted voting fusion strategy is employed to fusion the result of the two classifiers on each segment, and the whole file decision is made by selecting the most frequently labeled genre over all the segments. Experiments showed that the approach is effective. The fusion result gets 12.4% relative reduction in error rate compared to our baseline system.

For automatic language recognition, performance can be seriously degraded due to the transfer characteristics of the communication channel. Many methods are proposed to compensate the effect of the environment for better recognition results. In this paper, we propose a histogram based double Gaussian feature normalization method for robust language recognition. Compared with the baseline system, the proposed method achieves a relative error reduction of 17.4%, which shows advantages over other common feature normalization methods in language recognition systems.

Keyword spotting (KWS) is an essential technique for speech information retrieval. When doing offline keyword query on large volume spontaneous speech data, fast and accurate KWS methods are required. In this paper, a novel phone-state matrix based vocabulary-independent KWS method is proposed, which has merits of both hidden Markov model (HMM) based and lattice-based methods. Four KWS systems are compared in our experiments on conversational telephone speech test set. Result shows that compared to the high precision HMM-based KWS system the proposed phone-state matrix system has better equal-error-rate (EER) and false-alarm (FA) performance than the other two lattice-based systems.