Marcin D. Bugdol

Multimodal biometric system combining ECG and sound signals

In the paper a novel multimodal, behavioural biometric system that combines ECG and sound signal is described. The signal acquisition has been carried out in a small stress condition when the user has been asked to utter sounds at a given pitch. Thanks to the stimulations the ECG signal and especially the HR vary over time and these changes can be used to extract biometric features which represent an individual response to the stimuli. The proposed approach utilised popular statistical coefficients which are computationally effective and simple.

Dynamic time warping in phoneme modeling for fast pronunciation error detection

The presented paper describes a novel approach to the detection of pronunciation errors. It makes use of the modeling of well-pronounced and mispronounced phonemes by means of the Dynamic Time Warping (DTW) algorithm. Four approaches that make use of the DTW phoneme modeling were developed to detect pronunciation errors: Variations of the Word Structure (VoWS), Normalized Phoneme Distances Thresholding (NPDT), Furthest Segment Search (FSS) and Normalized Furthest Segment Search (NFSS). The performance evaluation of each module was carried out using a speech database of correctly and incorrectly pronounced words in the Polish language, with up to 10 patterns of every trained word from a set of 12 words having different phonetic structures. The performance of DTW modeling was compared to Hidden Markov Models (HMM) that were used for the same four approaches (VoWS, NPDT, FSS, NFSS). The average error rate (AER) was the lowest for DTW with NPDT (AER=0.287) and scored better than HMM with FSS (AER=0.473), which was the best result for HMM. The DTW modeling was faster than HMM for all four approaches. This technique can be used for computer-assisted pronunciation training systems that can work with a relatively small training speech corpus (less than 20 patterns per word) to support speech therapy at home.

An Idea of Human Voice Reaction Measurement System under the Aspect of Behavioral Biometric

A proposition of the human’s reaction measurement system focused on the voice is presented in the paper. The main idea of this system is to test the human behavior under the influence of sound and visual stimulations. In order to evaluate the human reaction, the signal of speech is examined. This signal is acquired in strict conditions using external device which have a high measurement accuracy.

Vowels in Speaker Verification

In the paper the possibility of using a single vowel in peoples’ identity verification process is described. The usage of sounds which period do not change over the whole duration time, allows applying Fourier series to its analysis. The amplitudes of consecutive harmonics as well as their frequencies are the basis of creating individual features’ vector. In order to record sounds of such characteristic, the acquisition process has been performed taking strict measurement’s assumption into consideration. The similarity measure of two vowels can be a coefficient which construction is based on the Euclidean distance with contribution of particular components.

Pronunciation Error Detection Using Dynamic Time Warping Algorithm

In the paper a pronunciation error detection method has been presented, wchich is based on word structural features. A lowcomplexity classifier has been proposed, that is not concentrated on a limited base of error patterns, but is flexible enough to find unspecified mispronunciations. Two classification variants using Dynamic Time Warping (DTW) algorithm has been tested on speech corpus containing recordings of 30 people.

Phase angles of sound as a biometric feature

The paper is a proposal of concept of highlighting the sound sources based on the phase shifts of successive harmonics in a single function course extracted from a articulation of single sound which frequency is stable in short-term. The function describing the single period of sound is expanded for the purpose of determining the harmonic series from −∞ to +∞. It is also an additional protection of sensitive personal data, because, based on such generated sound it is impossible to unambiguously recognize the voice of the individual using the human senses.

Driver biomedical support system

The article presents the concepts of drivers condition evaluation with the use of biomedical system. The described method is based on a thermometric measurements of selected, external areas of the body. In the paper summarized the results of correlating the temperature records with the driving style indicators. Additionally analysis of the vehicle cabin thermal distribution was made.

Human Activity Recognition Using Smartphone Sensors

In the paper a human activity recognition system has been presented based on the data gathered with the smartphone sensors. The acceleration, magnetic field and sound have been registered and four different activities of daily living has been recognized i.e. riding a bike, driving in a car, walking and sitting. Two version of Support Vector Machine (SVM) classifier have been employed and the obtained results are promising.

Strengthening a cryptographic system with behavioural biometric

The paper proposes a cryptographic system in which the encryption key has been associated both with the existing methods of verification of identity (knowledge and physical identifiers) as well as with biometrics. This combination allows spreading information about the marker and binding his relationship with the person entitled to use it. Biometric markers were extracted from the recorded voice and ECG signals which included the response of subjects to specific audio and visual stimuli. Results obtained from the experiments demonstrate the validity of presented concepts.

Boys’ Age Modeling Using Voice Features

The paper presents the results of boys’ age modeling on the basis of the features of their voice. The research group has been divided according to age and the threshold has been 14 years. 98 boys have been examined (57 aged less than 14 years, 41 aged 14 years or more). Voice data has been acquired and processed. The obtained coefficients have been subjected to Principal Component Analysis and then linear models have been built, estimeting the boys’ age. The obtained results are promising and are especially good in case of the group of younger boys, where the median absolute error has been less than 6 months and the median relative error has been equal to 2.1%.