Search

Over the last few years, i-vectors have been the state-of-the-art technique in speaker and language recognition. Recent advances in Deep Learning (DL) technology have improved the quality of i-vectors but the DL techniques in use are computationally expensive and need speaker or/and phonetic labels for the background data, which are not easily accessible in practice. On the other hand, the lack of speaker-labeled background data makes a big performance gap, in speaker recognition, between two well-known cosine and Probabilistic Linear Discriminant Analysis (PLDA) i-vector scoring techniques. It has recently been a challenge how to fill this gap without speaker labels, which are expensive in practice. Although some unsupervised clustering techniques are proposed to estimate the speaker labels, they cannot accurately estimate the labels. This thesis tries to solve the problems above by using the DL technology in different ways, without any need of speaker or phonetic labels. In order to fill the performance gap between cosine and PLDA scoring given unlabeled background data, we have proposed an impostor selection algorithm and a universal model adaptation process in a hybrid system based on Deep Belief Networks (DBNs) and Deep Neural Networks (DNNs) to discriminatively model each target speaker. In order to have more insight into the behavior of DL techniques in both single and multi-session speaker enrollment tasks, some experiments have been carried out in both scenarios. Experiments on the National Institute of Standard and Technology (NIST) 2014 i-vector challenge show that 46% of this performance gap, in terms of minDCF, is filled by the proposed DL-based system. Furthermore, the score combination of the proposed DL-based system and PLDA with estimated labels covers 79% of this gap. In the second line of the research, we have developed an efficient alternative vector representation of speech by keeping the computational cost as low as possible and avoiding phonet, En los últimos años, los i-vectores han sido la técnica de referencia en el reconocimiento de hablantes y de idioma. Los últimos avances en la tecnología de Aprendizaje Profundo (Deep Learning. DL) han mejorado la calidad de los i-vectores, pero las técnicas DL en uso son computacionalmente costosas y necesitan datos etiquetados para cada hablante y/o unidad fon ética, los cuales no son fácilmente accesibles en la práctica. La falta de datos etiquetados provoca una gran diferencia de los resultados en el reconocimiento de hablante con i-vectors entre las dos técnicas de evaluación más utilizados: distancia coseno y Análisis Lineal Discriminante Probabilístico (PLDA). Por el momento, sigue siendo un reto cómo reducir esta brecha sin disponer de las etiquetas de los hablantes, que son costosas de obtener. Aunque se han propuesto algunas técnicas de agrupamiento sin supervisión para estimar las etiquetas de los hablantes, no pueden estimar las etiquetas con precisión. Esta tesis trata de resolver los problemas mencionados usando la tecnología DL de diferentes maneras, sin necesidad de etiquetas de hablante o fon éticas. Con el fin de reducir la diferencia de resultados entre distancia coseno y PLDA a partir de datos no etiquetados, hemos propuesto un algoritmo selección de impostores y la adaptación a un modelo universal en un sistema hibrido basado en Deep Belief Networks (DBN) y Deep Neural Networks (DNN) para modelar a cada hablante objetivo de forma discriminativa. Con el fin de tener más información sobre el comportamiento de las técnicas DL en las tareas de identificación de hablante en una única sesión y en varias sesiones, se han llevado a cabo algunos experimentos en ambos escenarios. Los experimentos utilizando los datos del National Institute of Standard and Technology (NIST) 2014 i-vector Challenge muestran que el 46% de esta diferencia de resultados, en términos de minDCF, se reduce con el sistema propuesto basado en DL. Además, la combinación de evaluacio, Postprint (published version)

Over the last few years, i-vectors have been the state-of-the-art technique in speaker recognition. Recent advances in Deep Learning (DL) technology have improved the quality of i-vectors but the DL techniques in use are computationally expensive and need phonetically labeled background data. The aim of this work is to develop an efficient alternative vector representation of speech by keeping the computational cost as low as possible and avoiding phonetic labels, which are not always accessible. The proposed vectors will be based on both Gaussian Mixture Models (GMM) and Restricted Boltzmann Machines (RBM) and will be referred to as GMM–RBM vectors. The role of RBM is to learn the total speaker and session variability among background GMM supervectors. This RBM, which will be referred to as Universal RBM (URBM), will then be used to transform unseen supervectors to the proposed low dimensional vectors. The use of different activation functions for training the URBM and different transformation functions for extracting the proposed vectors are investigated. At the end, a variant of Rectified Linear Units (ReLU) which is referred to as variable ReLU (VReLU) is proposed. Experiments on the core test condition 5 of NIST SRE 2010 show that comparable results with conventional i-vectors are achieved with a clearly lower computational load in the vector extraction process., Peer Reviewed, Postprint (published version)

The lack of labeled background data makes a big performance gap between cosine and Probabilistic Linear Discriminant Analysis (PLDA) scoring baseline techniques for i-vectors in speaker recognition. Although there are some unsupervised clustering techniques to estimate the labels, they cannot accurately predict the true labels and they also assume that there are several samples from the same speaker in the background data that could not be true in reality. In this paper, the authors make use of Deep Learning (DL) to fill this performance gap given unlabeled background data. To this goal, the authors have proposed an impostor selection algorithm and a universal model adaptation process in a hybrid system based on deep belief networks and deep neural networks to discriminatively model each target speaker. In order to have more insight into the behavior of DL techniques in both single- and multisession speaker enrollment tasks, some experiments have been carried out in this paper in both scenarios. Experiments on National Institute of Standards and Technology 2014 i-vector challenge show that 46% of this performance gap, in terms of minimum of the decision cost function, is filled by the proposed DL-based system. Furthermore, the score combination of the proposed DL-based system and PLDA with estimated labels covers 79% of this gap., Peer Reviewed, Postprint (published version)

Restricted Boltzmann Machines (RBMs) have shown success in speaker recognition. In this paper, RBMs are investigated in a framework comprising a universal model training and model adaptation. Taking advantage of RBM unsupervised learning algorithm, a global model is trained based on all available background data. This general speaker-independent model, referred to as URBM, is further adapted to the data of a specific speaker to build speaker-dependent model. In order to show its effectiveness, we have applied this framework to two different tasks. It has been used to discriminatively model target and impostor spectral features for classification. It has been also utilized to produce a vector-based representation for speakers. This vector-based representation, similar to i-vector, can be further used for speaker recognition using either cosine scoring or Probabilistic Linear Discriminant Analysis (PLDA). The evaluation is performed on the core test condition of the NIST SRE 2006 database., Peer Reviewed, Postprint (author's final draft)

Restricted Boltzmann Machines (RBMs) have shown success in different stages of speaker recognition systems. In this paper, we propose a novel framework to produce a vector-based representation for each speaker, which will be referred to as RBM-vector. This new approach maps the speaker spectral features to a single fixed-dimensional vector carrying speaker-specific information. In this work, a global model, referred to as Universal RBM (URBM), is trained taking advantage of RBM unsupervised learning capabilities. Then, this URBM is adapted to the data of each speaker in the development, enrolment and evaluation datasets. The network connection weights of the adapted RBMs are further concatenated and subject to a whitening with dimension reduction stage to build the speaker vectors. The evaluation is performed on the core test condition of the NIST SRE 2006 database, and it is shown that RBM-vectors achieve 15% relative improvement in terms of EER compared to i-vectors using cosine scoring. The score fusion with i-vector attains more than 24% relative improvement. The interest of this result for score fusion yields on the fact that both vectors are produced in an unsupervised fashion and can be used instead of i-vector/PLDA approach, when no data label is available. Results obtained for RBM-vector/PLDA framework is comparable with the ones from i-vector/PLDA. Their score fusion achieves 14% relative improvement compared to i-vector/PLDA., Peer Reviewed, Postprint (published version)

This paper is focused on the application of the Language Identification (LID) technology for intelligent vehicles. We cope with short sentences or words spoken in moving cars in four languages: English, Spanish, German, and Finnish. As the response time of the LID system is crucial for user acceptance in this particular task, speech signals of different durations with total average of 3.8s are analyzed. In this paper, the authors propose the use of Deep Neural Networks (DNN) to model effectively the i-vector space of languages. Both raw i-vectors and session variability compensated i-vectors are evaluated as input vectors to DNNs. The performance of the proposed DNN architecture is compared with both conventional GMM-UBM and i-vector/LDA systems considering the effect of durations of signals. It is shown that the signals with durations between 2 and 3s meet the requirements of this application, i.e., high accuracy and fast decision, in which the proposed DNN architecture outperforms GMM-UBM and i-vector/LDA systems by 37% and 28%, respectively., Peer Reviewed, Postprint (published version)

The use of Restricted Boltzmann Machines (RBM) is proposed in this paper as a non-linear transformation of GMM supervectors for speaker recognition. It will be shown that the RBM transformation will increase the discrimination power of raw GMM supervectors for speaker recognition. The experimental results on the core test condition of the NIST SRE 2006 corpus show that the proposed RBM supervectors will achieve a comparable performance to i-vectors. Furthermore, the combination of RBM supevectors and i-vectors in the score level improves the performance of the i-vector approach by more than 10% in terms of EER., Peer Reviewed, Postprint (published version)

In this paper, we propose to discriminatively model target and impostor spectral features using Deep Belief Networks (DBNs) for speaker recognition. In the feature level, the number of impostor samples is considerably large compared to previous works based on i-vectors. Therefore, those i-vector based impostor selection algorithms are not computationally practical. On the other hand, the number of samples for each target speaker is different from one speaker to another which makes the training process more difficult. In this work, we take advantage of DBN unsupervised learning to train a global model, which will be referred to as Universal DBN (UDBN). Then we adapt this UDBN to the data of each target speaker. The evaluation is performed on the core test condition of the NIST SRE 2006 database and it is shown that the proposed architecture achieves more than 8% relative improvement in comparison to the conventional Multilayer Perceptron (MLP)., Peer Reviewed, Postprint (published version)

In this paper we propose an impostor selection method for a Deep Belief Network (DBN) based system which models i-vectors in a multi-session speaker verification task. In the proposed method, instead of choosing a fixed number of most informative impostors, a threshold is defined according to the frequencies of impostors. The selected impostors are then clustered and the centroids are considered as the final impostors for target speakers. The system first trains each target speaker unsupervisingly by an adaptation method and then models discriminatively each target speaker using the impostor centroids and target i-vectors. The evaluation is performed on the NIST 2014 i-vector challenge database and it is shown that the proposed DBN-based system achieves 23% relative improvement of minDCF over the baseline system in the challenge, Postprint (published version)

An effective global impostor selection method is proposed in this paper for discriminative Deep Belief Networks (DBN) in the context of a multi-session i-vector based speaker recognition. The proposed method is an iterative process in which in each iteration the whole impostor i-vector dataset is divided randomly into two subsets. The impostors in one subset which are closer to each impostor in another subset are selected and impostor frequencies are computed. At the end, those impostors with higher frequencies will be the global selected ones. They are then clustered and the centroids are considered as the final impostors for the DBN speaker models. The advantage of the proposed method is that in contrary to other similar approaches, only the background i-vector dataset is employed. The experimental results are performed on the NIST 2014 i-vector challenge dataset and it is shown that the proposed selection method improves the performance of the DBN-based system in terms of minDCF by 7% and the whole system outperforms the baseline in the challenge by more than 22% relative improvement., Peer Reviewed, Postprint (published version)

The acoustic environment of a typical neonatal intensive care unit (NICU) is very rich and may contain a large number of different sounds, which come either from the equipment or from the human activities taking place in it. There exists a medical concern about the effect of that acoustical environment on preterm infants, since loud sounds or particular sounds may be harmful for their further neurological development. In this work, first of all, an initial description of the acoustic characteristics of the NICU has been carried out using a set of diverse recordings produced with microphones placed both inside and outside an incubator. Then, the work has focused on detection of the most relevant types of sounds. In this paper, after describing the recorded database and the acoustic environment, preliminary experiments for detection of the acoustic alarms of devices are reported. The proposed detection system is based on Deep Belief Networks (DBN). The experimental results show that the DBN-based system is able to achieve better results than a baseline GMM-based system., Peer Reviewed, Postprint (published version)

The use of Deep Belief Networks (DBNs) is proposed in this paper to model discriminatively target and impostor i-vectors in a speaker verification task. The authors propose to adapt the network parameters of each speaker from a background model, which will be referred to as Universal DBN (UDBN). It is also suggested to backpropagate class errors up to only one layer for few iterations before to train the network. Additionally, an impostor selection method is introduced which helps the DBN to outperform the cosine distance classifier. The evaluation is performed on the core test condition of the NIST SRE 2006 corpora, and it is shown that 10% and 8% relative improvements of EER and minDCF can be achieved, respectively., Peer Reviewed, Postprint (published version)