Investigating in-domain data requirements for PLDA training

This paper analyzes the limitations upon the amount of in- domain (NIST SREs) data required for training a probabilistic linear discriminant analysis (PLDA) speaker verification system based on out-domain (Switchboard) total variability subspaces. By limiting the number of speakers, the number of sessions per speaker and the length of active speech per session available in the target domain for PLDA training, we investigated the relative effect of these three parameters on PLDA speaker verification performance in the NIST 2008 and NIST 2010 speaker recognition evaluation datasets. Experimental results indicate that while these parameters depend highly on each other, to beat out-domain PLDA training, more than 10 seconds of active speech should be available for at least 4 sessions/speaker for a minimum of 800 speakers. If further data is available, considerable improvement can be made over solely out-domain PLDA training.

Feature Selection for Pose Invariant Lip Biometrics

For the first time in this paper the authors present results showing the effect of out of plane speaker head pose variation on a lip biometric based speaker verification system. Using appearance DCT based features, they adopt a Mutual Information analysis technique to highlight the class discriminant DCT components most robust to changes in out of plane pose. Experiments are conducted using the initial phase of a new multi view Audio-Visual database designed for research and development of pose-invariant speech and speaker recognition. They show that verification performance can be improved by substituting higher order horizontal DCT components for vertical, particularly in the case of a train/test pose angle mismatch.

Robust Multimodal Person Identification With Limited Training Data

This paper presents a novel method of audio-visual feature-level fusion for person identification where both the speech and facial modalities may be corrupted, and there is a lack of prior knowledge about the corruption. Furthermore, we assume there are limited amount of training data for each modality (e.g., a short training speech segment and a single training facial image for each person). A new multimodal feature representation and a modified cosine similarity are introduced to combine and compare bimodal features with limited training data, as well as vastly differing data rates and feature sizes. Optimal feature selection and multicondition training are used to reduce the mismatch between training and testing, thereby making the system robust to unknown bimodal corruption. Experiments have been carried out on a bimodal dataset created from the SPIDRE speaker recognition database and AR face recognition database with variable noise corruption of speech and occlusion in the face images. The system's speaker identification performance on the SPIDRE database, and facial identification performance on the AR database, is comparable with the literature. Combining both modalities using the new method of multimodal fusion leads to significantly improved accuracy over the unimodal systems, even when both modalities have been corrupted. The new method also shows improved identification accuracy compared with the bimodal systems based on multicondition model training or missing-feature decoding alone.

Development of a Biometric Personal Authentication System Based on Fingerprint and Speech

Biometrics deals with the physiological and behavioral characteristics of an individual to establish identity. Fingerprint based authentication is the most advanced biometric authentication technology. The minutiae based fingerprint identification method offer reasonable identification rate. The feature minutiae map consists of about 70-100 minutia points and matching accuracy is dropping down while the size of database is growing up. Hence it is inevitable to make the size of the fingerprint feature code to be as smaller as possible so that identification may be much easier. In this research, a novel global singularity based fingerprint representation is proposed. Fingerprint baseline, which is the line between distal and intermediate phalangeal joint line in the fingerprint, is taken as the reference line. A polygon is formed with the singularities and the fingerprint baseline. The feature vectors are the polygonal angle, sides, area, type and the ridge counts in between the singularities. 100% recognition rate is achieved in this method. The method is compared with the conventional minutiae based recognition method in terms of computation time, receiver operator characteristics (ROC) and the feature vector length. Speech is a behavioural biometric modality and can be used for identification of a speaker. In this work, MFCC of text dependant speeches are computed and clustered using k-means algorithm. A backpropagation based Artificial Neural Network is trained to identify the clustered speech code. The performance of the neural network classifier is compared with the VQ based Euclidean minimum classifier. Biometric systems that use a single modality are usually affected by problems like noisy sensor data, non-universality and/or lack of distinctiveness of the biometric trait, unacceptable error rates, and spoof attacks. Multifinger feature level fusion based fingerprint recognition is developed and the performances are measured in terms of the ROC curve. Score level fusion of fingerprint and speech based recognition system is done and 100% accuracy is achieved for a considerable range of matching threshold

Improved Biometric Authentication System Using Multimodal Cue Integration

Biometrics is an efficient technology with great possibilities in the area of security system development for official and commercial applications. The biometrics has recently become a significant part of any efficient person authentication solution. The advantage of using biometric traits is that they cannot be stolen, shared or even forgotten. The thesis addresses one of the emerging topics in Authentication System, viz., the implementation of Improved Biometric Authentication System using Multimodal Cue Integration, as the operator assisted identification turns out to be tedious, laborious and time consuming. In order to derive the best performance for the authentication system, an appropriate feature selection criteria has been evolved. It has been seen that the selection of too many features lead to the deterioration in the authentication performance and efficiency. In the work reported in this thesis, various judiciously chosen components of the biometric traits and their feature vectors are used for realizing the newly proposed Biometric Authentication System using Multimodal Cue Integration. The feature vectors so generated from the noisy biometric traits is compared with the feature vectors available in the knowledge base and the most matching pattern is identified for the purpose of user authentication. In an attempt to improve the success rate of the Feature Vector based authentication system, the proposed system has been augmented with the user dependent weighted fusion technique.

Wavelet-based cepstrum calculation

In this paper we present a new wavelet-based algorithm for low-cost computation of the cepstrum. It can be used for real time precise pitch determination in automatic speech and speaker recognition systems. Many wavelet families are examined to determine the one that works best. The results confirm the efficacy and accuracy of the proposed technique for pitch extraction. (C) 2008 Elsevier B.V. All rights reserved.

On the determination of epsilon during discriminative GMM training

Discriminative training of Gaussian Mixture Models (GMMs) for speech or speaker recognition purposes is usually based on the gradient descent method, in which the iteration step-size, ε, uses to be defined experimentally. In this letter, we derive an equation to adaptively determine ε, by showing that the second-order Newton-Raphson iterative method to find roots of equations is equivalent to the gradient descent algorithm. © 2010 IEEE.

GMM-based classifiers for the automatic detection of obstructive sleep apnea

The aim of automatic pathological voice detection systems is to serve as tools, to medical specialists, for a more objective, less invasive and improved diagnosis of diseases. In this respect, the gold standard for those system include the usage of a optimized representation of the spectral envelope, either based on cepstral coefficients from the mel-scaled Fourier spectral envelope (Mel-Frequency Cepstral Coefficients) or from an all-pole estimation (Linear Prediction Coding Cepstral Coefficients) forcharacterization, and Gaussian Mixture Models for posterior classification. However, the study of recently proposed GMM-based classifiers as well as Nuisance mitigation techniques, such as those employed in speaker recognition, has not been widely considered inpathology detection labours. The present work aims at testing whether or not the employment of such speaker recognition tools might contribute to improve system performance in pathology detection systems, specifically in the automatic detection of Obstructive Sleep Apnea. The testing procedure employs an Obstructive Sleep Apnea database, in conjunction with GMM-based classifiers looking for a better performance. The results show that an improved performance might be obtained by using such approach.

Relevance of the glottal pulse and the vocal tract in gender detection

Gender detection is a very important objective to improve efficiency in tasks as speech or speaker recognition, among others. Traditionally gender detection has been focused on fundamental frequency (f0) and cepstral features derived from voiced segments of speech. The methodology presented here consists in obtaining uncorrelated glottal and vocal tract components which are parameterized as mel-frequency coefficients. K-fold and cross-validation using QDA and GMM classifiers showed that better detection rates are reached when glottal source and vocal tract parameters are used in a gender-balanced database of running speech from 340 speakers.

Desarrollo de una herramienta de análisis de la señal de voz

El habla es la principal herramienta de comunicación de la que dispone el ser humano que, no sólo le permite expresar su pensamiento y sus sentimientos sino que le distingue como individuo. El análisis de la señal de voz es fundamental para múltiples aplicaciones como pueden ser: síntesis y reconocimiento de habla, codificación, detección de patologías, identificación y reconocimiento de locutor… En el mercado se pueden encontrar herramientas comerciales o de libre distribución para realizar esta tarea. El objetivo de este Proyecto Fin de Grado es reunir varios algoritmos de análisis de la señal de voz en una única herramienta que se manejará a través de un entorno gráfico. Los algoritmos están siendo utilizados en el Grupo de investigación en Aplicaciones MultiMedia y Acústica de la Universidad Politécnica de Madrid para llevar a cabo su tarea investigadora y para ofertar talleres formativos a los alumnos de grado de la Escuela Técnica Superior de Ingeniería y Sistemas de Telecomunicación. Actualmente se ha encontrado alguna dificultad para poder aplicar los algoritmos ya que se han ido desarrollando a lo largo de varios años, por distintas personas y en distintos entornos de programación. Se han adaptado los programas existentes para generar una única herramienta en MATLAB que permite: . Detección de voz . Detección sordo/sonoro . Extracción y revisión manual de frecuencia fundamental de los sonidos sonoros . Extracción y revisión manual de formantes de los sonidos sonoros En todos los casos el usuario puede ajustar los parámetros de análisis y se ha mantenido y, en algunos casos, ampliado la funcionalidad de los algoritmos existentes. Los resultados del análisis se pueden manejar directamente en la aplicación o guardarse en un fichero. Por último se ha escrito el manual de usuario de la aplicación y se ha generado una aplicación independiente que puede instalarse y ejecutarse aunque no se disponga del software o de la versión adecuada de MATLAB. ABSTRACT. The speech is the main communication tool which has the human that as well as allowing to express his thoughts and feelings distinguishes him as an individual. The analysis of speech signal is essential for multiple applications such as: synthesis and recognition of speech, coding, detection of pathologies, identification and speaker recognition… In the market you can find commercial or open source tools to perform this task. The aim of this Final Degree Project is collect several algorithms of speech signal analysis in a single tool which will be managed through a graphical environment. These algorithms are being used in the research group Aplicaciones MultiMedia y Acústica at the Universidad Politécnica de Madrid to carry out its research work and to offer training workshops for students at the Escuela Técnica Superior de Ingeniería y Sistemas de Telecomunicación. Currently some difficulty has been found to be able to apply the algorithms as they have been developing over several years, by different people and in different programming environments. Existing programs have been adapted to generate a single tool in MATLAB that allows: . Voice Detection . Voice/Unvoice Detection . Extraction and manual review of fundamental frequency of voiced sounds . Extraction and manual review formant voiced sounds In all cases the user can adjust the scan settings, we have maintained and in some cases expanded the functionality of existing algorithms. The analysis results can be managed directly in the application or saved to a file. Finally we have written the application user’s manual and it has generated a standalone application that can be installed and run although the user does not have MATLAB software or the appropriate version.

Voice-processing technologies--their application in telecommunications.

As the telecommunications industry evolves over the next decade to provide the products and services that people will desire, several key technologies will become commonplace. Two of these, automatic speech recognition and text-to-speech synthesis, will provide users with more freedom on when, where, and how they access information. While these technologies are currently in their infancy, their capabilities are rapidly increasing and their deployment in today's telephone network is expanding. The economic impact of just one application, the automation of operator services, is well over $100 million per year. Yet there still are many technical challenges that must be resolved before these technologies can be deployed ubiquitously in products and services throughout the worldwide telephone network. These challenges include: (i) High level of accuracy. The technology must be perceived by the user as highly accurate, robust, and reliable. (ii) Easy to use. Speech is only one of several possible input/output modalities for conveying information between a human and a machine, much like a computer terminal or Touch-Tone pad on a telephone. It is not the final product. Therefore, speech technologies must be hidden from the user. That is, the burden of using the technology must be on the technology itself. (iii) Quick prototyping and development of new products and services. The technology must support the creation of new products and services based on speech in an efficient and timely fashion. In this paper I present a vision of the voice-processing industry with a focus on the areas with the broadest base of user penetration: speech recognition, text-to-speech synthesis, natural language processing, and speaker recognition technologies. The current and future applications of these technologies in the telecommunications industry will be examined in terms of their strengths, limitations, and the degree to which user needs have been or have yet to be met. Although noteworthy gains have been made in areas with potentially small user bases and in the more mature speech-coding technologies, these subjects are outside the scope of this paper.