Ewa Lukasik

Wavelet packets based features selection for voiceless plosives classification

There are contradictory reports on the usefulness of the wavelet packet transform (WPT) for feature extraction. This is mainly the case of signals of non-stationary character. In this paper we examine this tool for a category of short non-stationary speech signals, namely voiceless plosive consonants /p/, /t/, /k/. Three approaches to feature selection have been implemented: best basis search algorithm over the averaged wavelet packet coefficients of all data, local discriminant basis (LDB) algorithm, i.e. application of the best basis algorithm on the discriminant measure between coefficients in three classes and singular value decomposition (SVD) of the entropy matrices calculated from the wavelet packets for each class. The experiments conducted over the context independent plosives from speech database of Polish gave a classification rate higher for WPT based features than for traditional DFT based cepstral coefficients.

Attracting individuals and crowds with multimedia and a virtual artifact during a museum at night event

19 May 2012 was a very special date in Europe. Museums stayed open late in the night and invited visitors for free as part of the Museum at Night event. The challenge for museums at that night was to attract the crowds of visitors with the artifacts from museum collections. The paper presents a scenario of a Museum at Night presentation at the Museum of Musical Instruments in Poznan Poland that was devoted to a historical clavichord. The old instrument was exposed together with its fully operational virtual copy that could be played using a multi-touch screen. This kind of presentation, although very attractive for individual visitors, for groups gains additional value if it is accompanied by a multimedia presentation, music, film and oral explanation. A synergy of various stimuli proved to be compelling also during another event of this kind held at the University, called Scientists at Night.

Wavelet Packets Features Extraction and Selection for Discriminating Plucked Sounds of Violins

Plucked sounds of musical instruments from chordophones group are examples of non-stationary sounds having both tonal and transient Plucked souncharacter. The experiments presented in this paper had Plucked sounto answer to the question if the wavelet packet transform based strategy of features extraction and selection that proved useful in many other classification tasks will be also useful for distinguishing differences of sounds produced by master quality violins played pizzicato.

Skeleton-based audio envelope shape analysis

This paper presents a novel skeleton-based audio envelope representation and matching method for audio signal analysis. We propose using amplitude envelope in the time domain to represent and calculate the similarity between audio signals. To effectively describe the shape of each envelope, we employ the skeleton descriptor, namely Audio Skeleton, to integrate both geometrical and topological envelope features. Based on Audio Skeletons, the audio envelope matching can be substituted by searching for the correspondences of skeleton endpoints. Finally, the similarity between audio envelope shapes is calculated based on their correlated skeleton matching. Our main contributions include (i) the introduction of a skeleton-based audio envelope descriptor, (ii) a simple and efficient Audio Skeleton representation method and (iii) a fast skeleton pruning and matching algorithm.

Representing the evolving temporal envelope of musical instruments sounds using Computer Vision methods

This paper proposes application of shape retrieval method developed and used in the domain of Computer Vision to describe and to match the long-term temporal envelope of musical instruments sounds that are to be compared. To effectively describe each envelope, we employ the skeleton descriptor, namely Audio Skeleton, to integrate both geometrical and topological envelope features. Based on skeletons, the audio envelope matching can be substituted by searching for the correspondences of skeleton endpoints. Finally, the similarity of audio envelopes is calculated based on their correlated skeleton matching. Our main contributions include (i) the introduction of a novel audio envelope descriptor with skeleton and (ii) the efficient and fast audio skeleton pruning and matching algorithms. Our method is validated through the skeleton matching and audio retrieval experiments on AMATI violin sound dataset.

Mobile System for Optical Music Recognition and Music Sound Generation

The paper presents a mobile system for generating a melody based on a photo of a musical score. The client-server architecture was applied. The client role is designated to a mobile application responsible for taking a photo of a score, sending it to the server for further processing and playing mp3 file received from the server. The server role is to recognize notes from the image, generate mp3 file and send it to the client application. The key element of the system is the program realizing the algorithm of notes recognition. It is based on the decision trees and characteristics of the individual symbols extracted from the image. The system is implemented in the Windows Phone 8 framework and uses a cloud operating system Microsoft Azure. It enables easy archivization of photos, recognized notes in the Music XML format and generated mp3 files. An easy transition to other mobile operating systems is possible as well as processing multiple music collections scans.

Machine Learning Methods for Improving Gender Recognition from Stop Consonants

In the paper we present results of the extensive search for the best machine learning method enabling the recognition of data set with fuzzy boundaries between classes. The data under investigation were particular voiceless speech signals of non-stationary character: /p/, /t/ and /k/, carrying some information about gender of the speaker. The classification accuracy of 16 classifiers was compared, all working on the same attribute set consisting of cepstral coefficients. Reduced and extended (using feature construction methods) versions of attribute sets have also been considered. Surprisingly the best results have been observed for neural network and distance based classifiers and reached 87% of classification accuracy.