Search

Speech signals convey various pieces of information such as the identity of its speaker, the language spoken, and the linguistic information about the text being spoken, etc. In this paper, we extract information about the cell phones from their speech records by using mel-frequency cepstrum coefficients and identify their brands and models. Closed-set identification rates of 92.56% and 96.42% have been obtained on a set of 14 different cell phones in the experiments using vector quantization and support vector machine classifiers, respectively.

Remote patient tracking has recently gained increased attention, due to its lower cost and non-invasive nature. In this paper, the performance of Support Vector Machines (SVM), Least Square Support Vector Machines (LS-SVM), Multilayer Perceptron Neural Network (MLPNN), and General Regression Neural Network (GRNN) regression methods is studied in application to remote tracking of Parkinson's disease progression. Results indicate that the LS-SVM provides the best performance among the other three, and its performance is superior to that of the latest proposed regression method published in the literature.

Duration of training/test data has a considerable effect on the performance of a speaker recognition system. In this paper, we analyze the effect of training and test data duration and speaker gender on the performance of speaker recognition systems. Gaussian mixture models-universal background model (GMM-UBM), vector quantization-universal background model (VQ-UBM), support vector machines-generalized linear discriminant sequence kernel (SVM-GLDS) and support vector machines with GMM supervectors (GSV-SVM) are the classifiers we use for speaker recognition. Experimental results conducted on NIST 2002 and NIST 2005 speaker recognition evaluation (SRE) databases show that recognition performance breaks down when short utterances are used for training and testing independent from the recognizer (e.g. equal error rate (EER) reduces from 10.33% to 27.86% on NIST 2005) and GSV-SVM system yields higher EER than other methods in the case of using short utterances. It is also shown that recognition accuracy for male speakers are higher than female independent from database and classifier.

Bu çalışma, 25-29 Ağustos 2013 tarihleri arasında Lyon[Fransa]’da düzenlenen 14. Annual Conference of the International-Speech-Communication-Association (INTERSPEECH 2013)’da bildiri olarak sunulmuştur. We study the problem of vocal effort mismatch in speaker verification. Changes in speaker's vocal effort induce changes in fundamental frequency (F0) and formant structure which introduce unwanted intra-speaker variations to features. We compare seven alternative spectrum estimators in the context of melfrequency cepstral coefficient (MFCC) extraction for speaker verification. The compared variants include traditional FFT spectrum and six parametric all-pole models. Experimental results on the NIST 2010 speaker recognition evaluation (SRE) corpus utilizing both GMM-UBM and more recent GMM supervector classifier indicate that spectrum estimation has a considerable impact on speaker verification accuracy under mismatched vocal effort conditions. The highest recognition accuracy was achieved using a particular variant of temporally weighted all-pole model, stabilized weighted linear prediction (SWLP). Academy of Finland (253120) Int Speech Commun Assoc Europa org Amazon Microsoft Google TcL SYTRAL European Language Resources Assoc Ouaero Imaginove VOCAPIA res Acapela Speech ocean ALDEBARAN Orange Vecsys IBM Res Raytheon BBN Technol Voxygen

Bu çalışma, 25-29 Ağustos 2013 tarihlerinde Lyon[Fransa]'da düzenlenen 14. Annual Conference of the International Speech Communication Association [Interspeech 2013]'da bildiri olarak sunulmuştur. Popular features for speech processing, such as mel-frequency cepstral coefficients (MFCCs), are derived from the short-term magnitude spectrum, whereas the phase spectrum remains unused. While the common argument to use only the magnitude spectrum is that the human ear is phase-deaf, phase-based features have remained less explored due to additional signal processing difficulties they introduce. A useful representation of the phase is the group delay function, but its robust computation remains difficult. This paper advocates the use of group delay functions derived from parametric all-pole models instead of their direct computation from the discrete Fourier transform. Using a subset of the vocal effort data in the NIST 2010 speaker recognition evaluation (SRE) corpus, we show that group delay features derived via parametric all-pole models improve recognition accuracy, especially under high vocal effort. Additionally, the group delay features provide comparable or improved accuracy over conventional magnitude-based MFCC features. Thus, the use of group delay functions derived from all-pole models provide an effective way to utilize information from the phase spectrum of speech signals. Academy of Finland (253120) Int Speech Commun Association Amazon Microsoft Google TcL SYTRAL European Language Resources Association Ouaero Imaginove VOCAPIA Research Acapela Speech Ocean ALDEBARAN Orange Vecsys IBM Research Raytheon BBN Technology Voxygen

Bu çalışma, 26-31 Mayıs 2013 tarihleri arasında Vancouver[Kanada]’da düzenlenen IEEE International Conference on Acoustics, Speech, and Signal Processing (ICASSP)’da bildiri olarak sunulmuştur. Text-independent speaker identification is studied using neutral and shouted speech in Finnish to analyze the effect of vocal mode mismatch between training and test utterances. Standard mel-frequency cepstral coefficient (MFCC) features with Gaussian mixture model (GMM) recognizer are used for speaker identification. The results indicate that speaker identification accuracy reduces from perfect (100 %) to 8.71 % under vocal mode mismatch. Because of this dramatic degradation in recognition accuracy, we propose to use a joint density GMM mapping technique for compensating the MFCC features. This mapping is trained on a disjoint emotional speech corpus to create a completely speaker- and speech mode independent emotion-neutralizing mapping. As a result of the compensation, the 8.71 % identification accuracy increases to 32.00 % without degrading the non-mismatched train-test conditions much. Inst Elect & Elect Engineers Inst Elect & Elect Engineers Signal Proc Soc

Bu çalışma, 25-30 Mart 2012 tarihleri arasında Kyoto[Japonya]’da düzenlenen IEEE International Conference on Acoustics, Speech and Signal Processing’da bildiri olarak sunulmuştur. Different short-term spectrum estimators for speaker verification under additive noise are considered. Conventionally, mel-frequency cepstral coefficients (MFCCs) are computed from discrete Fourier transform (DFT) spectra of windowed speech frames. Recently, linear prediction (LP) and its temporally weighted variants have been substituted as the spectrum analysis method in speech and speaker recognition. In this paper, 12 different short-term spectrum estimation methods are compared for speaker verification under additive noise contamination. Experimental results conducted on NIST 2002 SRE show that the spectrum estimation method has a large effect on recognition performance and stabilized weighted LP (SWLP) and minimum variance distortionless response (MVDR) methods yield approximately 7 % and 8 % relative improvements over the standard DFT method at -10 dB SNR level of factory and babble noises, respectively in terms of equal error rate (EER). Inst Elect & Elect Engineers, Signal Processing Soc IEEE

Regularization of linear prediction based mel-frequency cepstral coefficient (MFCC) extraction in speaker verification is considered. Commonly, MFCCs are extracted from the discrete Fourier transform (DFT) spectrum of speech frames. In this paper, DFT spectrum estimate is replaced with the recently proposed regularized linear prediction (RLP) method. Regularization of temporally weighted variants, weighted LP (WLP) and stabilized WLP (SWLP) which have earlier shown success in speech and speaker recognition, is also introduced. A novel type of double autocorrelation (DAC) lag windowing is also proposed to enhance robustness. Experiments on the NIST 2002 corpus indicate that regularized all-pole methods (RLP, RWLP and RSWLP) yield large improvement on recognition accuracy under additive factory and babble noise conditions in terms of both equal error rate (EER) and minimum detection cost function (MinDCF). Academy of Finland European Commission - 132129 - 127345 European Commission - 238803

Bu çalışma, 29 Ağustos-2 Eylül 2011 tarihleri arasında Barselona[İspanya]'da düzenlenen 19. European Signal Processing Conference (Eusipco-2011)'de bildiri olarak sunulmuştur. The universal background model (UBM) based classifiers have recently been popular for speaker recognition. In this paper, we propose a dimension reduction method using local principal component analysis to improve the performance of speaker verification systems, where maximum a Posteriori (MAP) adapted vector quantization classifier (VQ-MAP or VQ-UBM) is employed. The proposed system first partitions the UBM data into disjoint regions (clusters) via conventional VQ algorithm and PCA is performed on the set of feature vectors in each region to obtain transformation matrix. Then, multiple speaker model is constructed using the set of transformed feature vectors closest to each cluster through MAP adaptation. Conducting experiments on NIST 2001 SRE, it is shown that transforming the data onto a lower dimensional space by the proposed method improves the recognition accuracy.

This paper evaluates the impact of three special forms of the Minkowski metric (Euclidean, City Block, and Chebychev distances) on the performance of the conventional vector quantization (VQ) and Gaussian mixture model (GMM) based closed-set text-independent speaker recognition systems, in terms of recognition rate and confidence on decisions. For the VQ based system, evaluations are carried out using the two most common clustering algorithms, LBG and K-means, and it is revealed which clustering algorithm and distance pair should be used to exploit the best attribute of both to achieve the best recognition rate for a given codebook size. In the case of GMM based system, we introduce the metrics into the GMM using a concatenation of the LBG and K-means algorithms in estimating the initial mean vectors, to which the system performance is sensitive, and explore their impact on system performance. We also make comparison of results obtained from evaluations on clean speech (TIMIT) and telephone speech databases (NTIMIT and NIST2001) with the modern classifiers VQ-UBM and GMM-UBM. It is found that there are cases where conventional VQ based system outperforms the modern systems. Moreover, the impact of distance metrics on the performance of the conventional and modern systems depends on the recognition task imposed (verification/identification).