Damian Cetnarowicz

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.

Fast Face Localisation Using AdaBoost Algorithm and Identification with Matrix Decomposition Methods

In this paper automatic recognition of faces in images is considered. This problem can be divided into to stages: localization and identification. Despite satisfactory accuracy of many existing face localisation algorithms, it is important to perform this operation efficiently, therefore a tool for fast face localisation with AdaBoost algorithm is presented. The face recognition is a difficult problem due to variations of illumination, facial expression, etc. In order to achieve good accuracy, proper features must be selected. A tool for face feature extraction (both holistic and localized) is presented.

Iterative SVD algorithm as a BSS solution

An important feature of an iterative SVD algorithm is presented, which, under certain circumstances, makes real-time calculations of approximate independent components possible. Presented experiments show that for mixtures of signals with low SNR, the algorithm is an efficient ICA solution with low computational cost and acceptable quality.

Gaussian mixture model speaker recognition system experiments with CORPORA database

In this paper current state-of-the-art speaker recognition system was tested with Polish Corpora database. System based on MFCC feature extraction method and Gaussian mixture model were used in several experiments. The minimum amount of data to train a reliable model has been determined. In the second experiment minimum length of a signal to recognize speakers has been determined. In the last part the speaker discriminative properties of Polish phonemes were investigated. The phonemes with the best speaker discriminative properties were determined.

Speaker tracking audio-video system

In this paper the system for tracking of a moving speaker is reported. The system utilizes audio visual information acquired from a camera and microphones. Using the camera a face of a speaker is detected. Two pairs of microphones allow for estimation of direction of sound arrival using general cross-correlation method. The preliminary results of experiments are described.