http://cs.uef.fi/sipu/pub/SRE-review-Kinnunen-and-Li.pdf

An overview of text-independent speaker recognition: From features to supervectors

This chapter introduces the subject of statistical pattern recognition (SPR). It starts by considering how features are defined and emphasizes that the nearest neighbor algorithm achieves error rates comparable with those of an ideal Bayes’ classifier. The concepts of an optimal number of features, representativeness of the training data, and the need to avoid overfitting to the training data are stressed. The chapter shows that methods such as the support vector machine and artificial neural networks are subject to these same training limitations, although each has its advantages. For neural networks, the multilayer perceptron architecture and back-propagation algorithm are described. The chapter distinguishes between supervised and unsupervised learning, demonstrating the advantages of the latter and showing how methods such as clustering and principal components analysis fit into the SPR framework. The chapter also defines the receiver operating characteristic, which allows an optimum balance between false positives and false negatives to be achieved.

Statistical Pattern Recognition

[서평]「Digital Video Processing」

Fusing the advantages of multiple acoustic features is important for the robustness of voice activity detection (VAD). Recently, the machine-learning-based VADs have shown a superiority to traditional VADs on multiple feature fusion tasks. However, existing machine-learning-based VADs only utilize shallow models, which cannot explore the underlying manifold of the features. In this paper, we propose to fuse multiple features via a deep model, called deep belief network (DBN). DBN is a powerful hierarchical generative model for feature extraction. It can describe highly variant functions and discover the manifold of the features. We take the multiple serially-concatenated features as the input layer of DBN, and then extract a new feature by transferring these features through multiple nonlinear hidden layers. Finally, we predict the class of the new feature by a linear classifier. We further analyze that even a single-hidden-layer-based belief network is as powerful as the state-of-the-art models in the machine-learning-based VADs. In our empirical comparison, ten common features are used for performance analysis. Extensive experimental results on the AURORA2 corpus show that the DBN-based VAD not only outperforms eleven referenced VADs, but also can meet the real-time detection demand of VAD. The results also show that the DBN-based VAD can fuse the advantages of multiple features effectively.

http://fileserv.wcl.ee.upatras.gr/projectm/sites/default/files/06362186.pdf

Deep Belief Networks Based Voice Activity Detection

An important drawback affecting most of the speech processing systems is the environmental noise and its harmful effect on the system performance. Examples of such systems are the new wireless communications voice services or digital hearing aid devices. In speech recognition, there are still technical barriers inhibiting such systems from meeting the demands of modern applications. Numerous noise reduction techniques have been developed to palliate the effect of the noise on the system performance and often require an estimate of the noise statistics obtained by means of a precise voice activity detector (VAD). Speech/non-speech detection is an unsolved problem in speech processing and affects numerous applications including robust speech recognition (Karray and Marting, 2003; Ramirez et al. 2003), discontinuous transmission (ITU, 1996; ETSI, 1999), real-time speech transmission on the Internet (Sangwan et al., 2002) or combined noise reduction and echo cancellation schemes in the context of telephony (Basbug et al., 2004; Gustafsson et al., 2002). The speech/non-speech classification task is not as trivial as it appears, and most of the VAD algorithms fail when the level of background noise increases. During the last decade, numerous researchers have developed different strategies for detecting speech on a noisy signal (Sohn et al., 1999; Cho and Kondoz, 2001; Gazor and Zhang, 2003, Armani et al., 2003) and have evaluated the influence of the VAD effectiveness on the performance of speech processing systems (Bouquin-Jeannes and Faucon, 1995). Most of the approaches have focussed on the development of robust algorithms with special attention being paid to the derivation and study of noise robust features and decision rules (Woo et al., 2000; Li et al., 2002; Marzinzik and Kollmeier, 2002). The different VAD methods include those based on energy thresholds (Woo et al., 2000), pitch detection (Chengalvarayan, 1999), spectrum analysis (Marzinzik and Kollmeier, 2002), zero-crossing rate (ITU, 1996), periodicity measure (Tucker, 1992), higher order statistics in the LPC residual domain (Nemer et al., 2001) or combinations of different features (ITU, 1993; ETSI, 1999; Tanyer and Ozer, 2000). This chapter shows a comprehensive approximation to the main challenges in voice activity detection, the different solutions that have been reported in a complete review of the state of the art and the evaluation frameworks that are normally used. The application of VADs for speech coding, speech enhancement and robust speech recognition systems is shown and discussed. Three different VAD methods are described and compared to standardized and

/pdf/voice-activity-detection-fundamentals-and-speech-recognition-11gtvo0nf8.pdf

Voice Activity Detection. Fundamentals and Speech Recognition System Robustness

Efficient voice activity detection algorithms using long-term speech information

This paper introduces recurrent neural networks (RNN), long short-term memory (LSTM), and gated recurrent unit (GRU) to detect and classify continuous sequences of volcano-seismic events at the Deception Island Volcano, Antarctica. A representative data set containing volcano-tectonic earthquakes, long-period events, volcanic tremors, and hybrid events was used to train these models. Experimental results show that RNN, LSTM, and GRU can exploit temporal and frequency information from continuous seismic data, attaining close to 90%, 94%, and 92% events correctly detected and classified. A second experiment is presented in this paper. The architectures described above, trained with data from campaigns of seismic records obtained in 1995–2002, have been tested with data from the recent seismic survey performed at the Deception Island Volcano in 2016–2017 by the Spanish Antarctic scientific campaign. Despite the variations in the geophysical properties of the seismic events within the volcano across eruptive periods, results provide good generalization accuracy. This result expands the possibilities of RNNs for real-time monitoring of volcanic activity, even if seismic sources change over time.

Detection and Classification of Continuous Volcano-Seismic Signals With Recurrent Neural Networks

Detection and classification of the different seismic events are important tasks in volcanological observatories. Trying to make these an automatic process is fundamental for the volcanological community. It is crucial to choose how the seismic signal is represented in terms of parameters or features useful for dealing with the automatic classification problem, since the number and type of parameters could be really large leading to the curse of dimensionality issue. Machine learning theory establishes that in order to build a classifier from a labeled database, there should be a compromise between the complexity of the classifier and the size of the database. Since generating a manually labeled database is a tedious work performed by specialists in volcanology, the size of the databases limits the complexity of the classification systems built by them. On the other hand, if the databases could be represented by a reduced, but relevant, number of features, the complexity of the classifier would be simplified. In order to study the problem just described, this paper performs a comparative study of different classical techniques of dimensionality reduction (DR) of the feature set. The algorithms implemented include feature selection techniques as wrappers and filters and methods which directly transform the original feature space into another with lower dimension. All algorithms have been tested using an automatic classification system of volcano-seismic events. The best results have been obtained with the discriminative feature selection (DFS) algorithm which belongs to the set of wrapper methods.

A Comparative Study of Dimensionality Reduction Algorithms Applied to Volcano-Seismic Signals

This paper shows an efficient voice activity detector (VAD) that is based on the estimation of the long-term spectral diver- gence (LTSD) between noise and speech periods. The proposed method decomposes the input signal into overlapped speech frames, uses a sliding window to compute the long-term spec- tral envelope and measures the speech/non-speech LTSD, thus yielding a high discriminating decision rule and minimizing the average number of decision errors. In order to increase non- speech detection accuracy, the decision threshold is adapted to the measured noise energy while a controlled hang-over is ac- tivated only when the observed signal-to-noise ratio (SNR) is low. An exhaustive analysis of the proposed VAD is carried out using the AURORA TIdigits and SpeechDat-Car (SDC) databases. The proposed VAD is compared to the most com- monly used ones in the field in terms of speech/non-speech detection and recognition performance. Experimental results demonstrate a sustained advantage over G.729, AMR and AFE VADs.

/pdf/a-new-adaptive-long-term-spectral-estimation-voice-activity-1ds8f2rv5u.pdf

A New Adaptive Long-Term Spectral Estimation Voice Activity Detector

This work is mainly focused on showing experimental results using a combination of two methods for noise compensation which are shown to be complementary: classical spectral subtraction algorithm and histogram equalization. While spectral subtraction is focused on the reduction of the additive noise in the spectral domain, histogram equalization is applied in the cepstral domain to compensate the remaining non-linear effects associated to channel distortion and additive noise. The estimation of the noise spectrum for the spectral subtraction method relies on a new algorithm for speech / non-speech detection (SND) based on order statistics. This SND classification is also used for dropping long speech pauses. Results on Aurora 2 and Aurora 3 are reported.

/pdf/feature-extraction-combining-spectral-noise-reduction-and-4c3dy2r7sq.pdf

M. Carmen Benítez

Papers

Efficient voice activity detection algorithms using long-term speech information

Detection and Classification of Continuous Volcano-Seismic Signals With Recurrent Neural Networks

A Comparative Study of Dimensionality Reduction Algorithms Applied to Volcano-Seismic Signals

A New Adaptive Long-Term Spectral Estimation Voice Activity Detector

Feature extraction combining spectral noise reduction and cepstral histogram equalization for robust ASR.