Gérard Chollet

BIOMET: a multimodal person authentication database including face, voice, fingerprint, hand and signature modalities

Information technology innovations involve a constant evolution of man-machine interaction modes. Automated authentication of people could be used to better adapt the machine to the user. Security can also be enhanced through a better people authentication. Biometrics appears as a promising tool in these two situations. Different modalities can be envisaged, such as: fingerprint, human face images, hand shape, voice, handwritten signature... In order to take advantage of the particularities of each modality, and to improve the performance of a person authentication system, multimodality can be applied. This motivated the recording of BIOMET, a biometric database with five different modalities, including face, voice, fingerprint, hand and signature data. In this paper, the BIOMET multimodal database for person authentication is described. Details about the acquisition protocols of each modality are given. Preliminary monomodal verification results, obtained on a subcorpus of the BIOMET fingerprint data, are also presented.

Confidence measures for keyword spotting using support vector machines

Support vector machines (SVM) is a new and very promising classification technique developed from the theory of structural risk minimisation. We propose an alternative out-of-vocabulary word detection method relying on confidence measures and support vector machines. Confidence measures are computed from phone level information provided by a hidden Markov model (HMM) based speech recognizer. We use three kinds of average techniques as arithmetic, geometric and harmonic averages to compute a confidence measure for each word. The acceptance/rejection decision of a word is based on the confidence feature vector which is processed by a SVM classifier. The performance of the proposed SVM classifier is compared with methods based on the averaging of confidence measures.

Nonlinear Speech Modeling and Applications

Dealing with Nonlinearities in Speech Signals.- Some Notes on Nonlinearities of Speech.- Nonlinear Speech Processing: Overview and Possibilities in Speech Coding.- Signal Processing in a Nonlinear, NonGaussian, and Nonstationary World.- Acoustic-to-Articulatory Modeling of Speech Phenomena.- The Analysis of Voice Quality in Speech Processing.- Identification of Nonlinear Oscillator Models for Speech Analysis and Synthesis.- Speech Modelling Based on Acoustic-to-Articulatory Mapping.- Data Driven and Speech Processing Algorithms.- Underdetermined Blind Separation of Speech Signals with Delays in Different Time-Frequency Domains.- Data Driven Approaches to Speech and Language Processing.- Cepstrum-Based Harmonics-to-Noise Ratio Measurement in Voiced Speech.- Predictive Connectionist Approach to Speech Recognition.- Modeling Speech Based on Harmonic Plus Noise Models.- Algorithms and Models Based on Speech Perception Mechanisms.- Text Independent Methods for Speech Segmentation.- Nonlinear Adaptive Speech Enhancement Inspired by Early Auditory Processing.- Perceptive, Non-linear Speech Processing and Spiking Neural Networks.- Task Oriented Speech Applications.- An Algorithm to Estimate Anticausal Glottal Flow Component from Speech Signals.- Non-linear Speech Feature Extraction for Phoneme Classification and Speaker Recognition.- Segmental Scores Fusion for ALISP-Based GMM Text-Independent Speaker Verification.- On the Usefulness of Almost-Redundant Information for Pattern Recognition.- An Audio-Visual Imposture Scenario by Talking Face Animation.- Cryptographic-Speech-Key Generation Using the SVM Technique over the lp-Cepstral Speech Space.- Nonlinear Speech Features for the Objective Detection of Discontinuities in Concatenative Speech Synthesis.- Signal Sparsity Enhancement Through Wavelet Transforms in Underdetermined BSS.- A Quantitative Evaluation of a Bio-inspired Sound Segregation Technique for Two- and Three-Source Mixtures.- Application of Symbolic Machine Learning to Audio Signal Segmentation.- Analysis of an Infant Cry Recognizer for the Early Identification of Pathologies.- Graphical Models for Text-Independent Speaker Verification.- An Application to Acquire Audio Signals with ChicoPlus Hardware.- Speech Identity Conversion.- Robust Speech Enhancement Based on NPHMM Under Unknown Noise.

Text-independent Speaker Verification

In this chapter, an overview of text-independent speaker verification is given first. Then, recent developments needed to reach state-of-the-art performances using low-level (acoustic) features as well as how to use complementary high-level information, are presented. The most relevant speaker verification evaluation campaigns and databases are also summarized. The BioSecure benchmarking framework for speaker verification using open-source state-of-the-art algorithms, well-known databases, and reference protocols is presented after. It is also shown how to reach state-of-the-art performances using open-source software with a case study example on the National Institute of Standards and Technology 2005 Speaker Evaluation data (NIST’2005 SRE). The examples of key factors influencing the performances of speaker verification experiments on the NIST’2005 evaluation data are grouped in three parts. The first set of experiments is related to the importance of front-end processing and data selection to fine-tune the acoustic Gaussian Mixture systems. The second set of experiments illustrates the importance of speaker and session variability modeling methods in order to cope with mismatched enrollment/test conditions. The third series of experiments demonstrates the usefulness of data-driven speech segmentation methods for extracting complementary high-level information. The chapter ends with conclusions and perspectives.

Coupling Context Awareness and Multimodality in Smart Homes Concept

Development of smart home technologies dedicated to people with disabilities provides a challenge in determining accurate requirements and needs in dynamic situations. In this paper we describe the integration of context awareness and multimodal functionalities in a smart environment. We outline how to optimize user comfort and capabilities. Considering the wide range of user types and preferences and the dynamic system environment created by constant introduction of new product and new context. By taking the environmental information provided by the environment, user profile and preferences, context awareness promises easier interaction and new possibilities such as predictive tasks automatically and adapting new situations to user interface. Multimodality permits in one hand to facilitate accessibility to a wide range of users, and the other hand to offer innovative control method of complex systems. In this paper we present our approach for coupling context awareness and multimodality concepts.

Combining methods to improve speaker verification decision

This paper describes how the combination of speaker verification algorithms with a priori decision thresholds can improve the overall robustness of a real application. The evaluation is performed in the context of a field application where each client is verified from a seven-digit personal identification number (PIN code). This paper demonstrates that it is possible to increase the global performance of the system by combining the results of several algorithms.

New cepstral representation using wavelet analysis and spectral transformation for robust speech recognition

The goal is to improve the speech recognition rate by optimisation of mel frequency cepstral coefficients (MFCCs): modifications concern the time-frequency representations used to estimate these coefficients. There are many ways to obtain a spectrum out of a signal which differ in the method itself (Fourier, wavelets,...), and in the normalisation. We show that we can obtain noise resistant cepstral coefficients, for speaker independent connected word recognition. The recognition system is based on a continuous whole word hidden Markov model. An error reduction rate of approximately 50% is achieved with word models.

Advances in Very Low Bit Rate Speech Coding Using Recognition and Synthesis Techniques

ALISP (Automatic Language Independent Speech Processing) units are an alternative concept to using phoneme-derived units in speech processing. This article describes advances in very low bit rate coding using ALISP units. Results of speaker-independent experiments are reported and speaker clustering using vector quantization is proposed. The improvements of speech re-synthesis using Harmonic Noise Model and dynamic selection of units are discussed.

Neural net approaches to speaker verification: comparison with second order statistic measures

The non-supervised self organizing map of Kohonen (SOM), the supervised learning vector quantization algorithm (LVQ3), and a method based on second-order statistical measures (SOSM) were adapted, evaluated and compared for speaker verification on 57 speakers of a POLYPHONE-like data base. The SOM and LVQ3 were trained by codebooks with 32 and 256 codes and two statistical measures; one without weighting (SOSM1) and another with weighting (SOSM2) were implemented. As the decision criterion, the equal error rate (EER) and best match decision rule (BMDR) were employed and evaluated. The weighted linear predictive cepstrum coefficients (LPCC) and the /spl Delta/LPCC were used jointly as two kinds of spectral speech representations in a single vector as distinctive features. The LVQ3 demonstrates a performance advantage over SOM. This is due to the fact that the LVQ3 allows the long-term fine-tuning of an interested target codebook using speech data from a client and other speakers, whereas the SOM only uses data from the client. The SOSM performs better than the SOM and the LVQ3 for long test utterances, while for short test utterances the LVQ is the best method among the methods studied.

Improving the performance of a keyword spotting system by using support vector machines

Support vector machines (SVM) represent a new approach to pattern classification developed from the theory of structural risk minimisation. In this paper, we propose an investigation into the application of SVM to the confidence measurement problem in speech recognition. Confidence measures are computed using the phone level information provided by a hidden Markov model (HMM) based speech recognizer. We use three kinds of average techniques as arithmetic, geometric and harmonic averages in order to compute a confidence measure for each word. The acceptance/rejection decision for a given word is based on the confidence feature vector which is processed by a SVM classifier. The performance of the proposed SVM classifier is compared with methods based on the averaging of phone confidence measures.