Adam Dabrowski

Influence of low resolution of images on reliability of face detection and recognition

In this paper we analyze reliability of the real-time system for face detection and recognition from low-resolution images, e.g., from video monitoring images. First, we briefly describe main features of the standards for biometric face images. Available scientific databases have been checked for compliance with these biometric standards. During the research we have considered both the correctness of extraction (location) of the face from the image as well as the correctness of the identification (based on the eigenfaces approach). To the tests we have used the face databases that allow to study tolerance to illumination and face positions. We have compared various face detection techniques and analyzed minimum requirements for the resolution of facial images. Finally, an influence of the resolution reduction on the FAR/FRR of the recognition is presented.

Detection of GSM speech coding for telephone call classification and automatic speaker recognition

This paper presents results of our research on automatic telephone call speaker recognition. A significant improvement of recognition quality can be observed if we use the knowledge on how the original speech was coded, e.g., using GSM or PCM telephone standard. On this basis we can improve the quality of the speaker models. In order to realize such a speech analyzer we propose to analyze the mean square error of the transcoded speech.

Face Recognition from Low Resolution Images

This paper describes an analysis of the real-time system for face recognition from video monitoring images. First, we briefly describe main features of the standards for biometric face images. Available scientific databases have been checked for compliance with these biometric standards. Next, we concentrate on the analysis of the prepared face recognition application based on the eigenface approach. Finally, results of our face recognition experiments with images of reduced resolution are presented. It turned out that the proposed and tested algorithm is quite resistant to changing the resolution. The recognition results are acceptable even for low-resolution images (16×20 pixels).

Generalized convolution as a tool for the multi-dimensional filtering tasks

In this paper the so-called generalized convolution, being in fact an adequate adaptation of the well known circular convolution concept to any invertible block-transform, is proposed, developed, and analysed. First the proposed idea is summarized for a one-dimensional case. Then it is extended to multidimensional problems. The presented generalized convolution concept is based on the earlier A-convolution. This idea is recalled at the beginning and a set of techniques for studying the dependence of the respective coefficients on the arbitrary transform, is suggested. The generalized convolution matrix, being an extension of that for the circular convolution, is introduced and adapted to an arbitrary invertible transform. The filtering problem is then defined and presented in the transform domain. The multidimensional analysis starts with two-dimensional problems, then it is continued with formulas for multidimensional filtering tasks. The paper is illustrated with examples computed for twenty carefully selected transforms. Among them are Haar, Hadamard, Hartley, Karhunen-Loeve and a family of 16 discrete trigonometric transforms.

Analysis of speech separation for ASR systems

This work presents some aspects concentrated on improving quality of the voice input signal for automatic speech recognition (ASR) systems. It is clear that proper separation of the desired voice from disturbances improves the recognition performance. We have tested the independent component analysis (ICA) infomax algorithm applied to the separation of isolated words. Detection of Polish words (commands used for the remote voice control of a robot) was tested using the dynamic time warping (DTW) approach.

Software defined radio - design and implementation of complete platform

In this paper, a complete architecture of the software defined radio receiver is proposed. First, a precise definition and the discussion of the problem is given. Next, an outline of the system topology is proposed, together with the requirements for each component of the system. Finally, a custom implementation is presented, including specifications, design considerations, and simulations.

Bluetooth technology applications dedicated to supporting blind and hearing as well as speech handicapped people

In this paper a concept of using the Bluetooth wireless telecommunication system for the construction of devices supporting blind, hearing impaired, and speech handicapped people is presented. People of the third considered class can be mute from infancy or their speech organs can be disordered after larynx surgical operations. Various possibilities of the Bluetooth technology applications in supporting patients with the disfunctions mentioned above are discussed and developed. One of the suggested solutions - a prototype speech-aid system - is described in detail

Speaker recognition based on multilevel speech signal analysis on Polish corpus

This article deals with a new approach to the text-independent speaker verification task. It is namely proposed to combine spectral and the so-called high-level features (prosodic, articulatory, and lexical) in order to increase accuracy of speaker verification. The presented experiments were performed using a Polish language corpus developed by the authors, the so-called PUEPS corpus. It contains semi-spontaneous telephone conversations (acted emergency telephone notifications) recorded in laboratory conditions. As the Polish language is under resourced and the PUEPS corpus is relatively small, in this case a new approach is needed, other than these well known from NIST (National Institute of Standards and Technology) evaluations. The authors proposed to use the fast scoring instead of more complex classifiers and the AdaBoost (adaptive boosting) algorithm for features combination. Combination of features resulted in the equal error rate (EER) reduction for various SNR (signal-to-noise ratio) conditions. Additionally, score normalization methods were evaluated. It was shown that significant benefits can be obtained using the z-norm2 method.

Improving of Speaker Identification from Mobile Telephone Calls

The paper examines issues related to proper selection of models used for quick speaker recognition based on short recordings of mobile telephone conversations. A knowledge of the encoder type used during the transmission of speech allows to apply an appropriate model that takes specific characteristics of the encoder into account: full rate (FR), half rate (HR), enhanced full rate (EFR) and adaptive multi-rate (AMR). We analyse both proper model selection and automatic silence removal. Analysis of time of processing is also a part of this study.

Proposal of VLIW Architecture for Application Specific Processors with Built-in-Self-Repair facility via Variable Accuracy Arithmetic

A new proposal of a very long instruction word (VLIW) architecture for application specific processors with the built-in-self-repair (BISR) facility realized by means of the variable accuracy arithmetic has been described in this paper. The proposed idea, which is particularly interesting for the portable embedded systems, is based on a novel two accumulator model, which is used in order to obtain high accuracy of the result of a series of floating-point additions. The goal of the proposed approach consists in ensuring the functionality of the system even in case of permanent faults of particular building blocks, e.g., of one of the floating point adders. Instead of decreasing the performance of the system or providing redundant hardware, as it is typically done in other approaches, in our present concept merely the accuracy of the floating point accumulation is reduced. Therefore, although the quality of service is reduced, the performance and the overall functionality of the system may remain unchanged. By this means we can also control the power consumption, saving the power with the cost of the somehow reduced signal processing accuracy.