Categorization of Stringed Instruments with Multifractal Detrended Fluctuation Analysis

📝 Abstract

Categorization is crucial for content description in archiving of music signals. On many occasions, human brain fails to classify the instruments properly just by listening to their sounds which is evident from the human response data collected during our experiment. Some previous attempts to categorize several musical instruments using various linear analysis methods required a number of parameters to be determined. In this work, we attempted to categorize a number of string instruments according to their mode of playing using latest-state-of-the-art robust non-linear methods. For this, 30 second sound signals of 26 different string instruments from all over the world were analyzed with the help of non linear multifractal analysis (MFDFA) technique. The spectral width obtained from the MFDFA method gives an estimate of the complexity of the signal. From the variation of spectral width, we observed distinct clustering among the string instruments according to their mode of playing. Also there is an indication that similarity in the structural configuration of the instruments is playing a major role in the clustering of their spectral width. The observations and implications are discussed in detail.

💡 Analysis

Categorization is crucial for content description in archiving of music signals. On many occasions, human brain fails to classify the instruments properly just by listening to their sounds which is evident from the human response data collected during our experiment. Some previous attempts to categorize several musical instruments using various linear analysis methods required a number of parameters to be determined. In this work, we attempted to categorize a number of string instruments according to their mode of playing using latest-state-of-the-art robust non-linear methods. For this, 30 second sound signals of 26 different string instruments from all over the world were analyzed with the help of non linear multifractal analysis (MFDFA) technique. The spectral width obtained from the MFDFA method gives an estimate of the complexity of the signal. From the variation of spectral width, we observed distinct clustering among the string instruments according to their mode of playing. Also there is an indication that similarity in the structural configuration of the instruments is playing a major role in the clustering of their spectral width. The observations and implications are discussed in detail.

📄 Content

CATEGORIZATION OF STRINGED INSTRUMENTS WITH MULTIFRACTAL DETRENDED FLUCTUATION ANALYSIS

Archi Banerjee*, Shankha Sanyal, Tarit Guhathakurata, Ranjan Sengupta and Dipak Ghosh

Sir C.V. Raman Centre for Physics and Music Jadavpur University, Kolkata: 700032

*archibanerjee7@gmail.com

• Corresponding Author

ABSTRACT
Categorization is crucial for content description in archiving of music signals. On many occasions, human brain fails to classify the instruments properly just by listening to their sounds which is evident from the human response data collected during our experiment. Some previous attempts to categorize several musical instruments using various linear analysis methods required a number of parameters to be determined. In this work, we attempted to categorize a number of string instruments according to their mode of playing using latest-state-of-the-art robust non-linear methods. For this, 30 second sound signals of 26 different string instruments from all over the world were analyzed with the help of non linear multifractal analysis (MFDFA) technique. The spectral width obtained from the MFDFA method gives an estimate of the complexity of the signal. From the variation of spectral width, we observed distinct clustering among the string instruments according to their mode of playing. Also there is an indication that similarity in the structural configuration of the instruments is playing a major role in the clustering of their spectral width. The observations and implications are discussed in detail.

Keywords: String Instruments, Categorization, Fractal Analysis, MFDFA, Spectral Width

INTRODUCTION Classification is one of the processes involved in audio content description. Audio signals can be classified according to miscellaneous criteria viz. speech, music, sound effects (or noises). Usually, music streams are broadly classified according to genre, player, mood, or instrumentation [1]. Some of the most important instruments in the history of music have been stringed instruments, which range from early to modern forms of the violin and the guitar, through to contemporary experiments with amplification and electric or digital recording. Stringed instruments are often said to belong to different categories based on their timbre or how they produce sound. The string may be struck, plucked, rubbed (bowed), or, occasionally, blown (by the wind); in each case the effect is to displace the string from its normal position of rest and to cause it to vibrate in complex patterns. All these factors characterize the “timbre” of a stringed instrument. In a sense, timbre is everything that lets us distinguish one instrument from another. For example, a trumpet has a brash timbre compared with the smoother timbre of a saxophone. A violin and viola have very similar timbres (which is the reason we might struggle to distinguish them by ear, but can be distinguished through robust analysis). The timbre of a string instrument is completely different from a keyed or wind instrument. Timbre is a combination of many factors. It includes tone but also aspects like how suddenly or smoothly the notes start and end, the number and strength of harmonics in the sound, and how the sound varies over time. No system of classification can adequately categorize the interactions of natural material, craftsmanship, and exuberant imagination that produced an endless variety of stringed instruments. We envisaged developing a scientific technique involving only one or two parameters with the help of which we can automatically ascertain that a particular stringed instrument belongs to a certain family of stringed instruments. This automated technique also clearly characterizes groups of instruments which seem to hear similar to human ear but is distinguishable with the method developed. Mandelbrot [2] demonstrated how nature contains structures (e.g., mountains, coastlines, the structures of plants), which could be described by fractals and suggested that fractal theory could be used in order to understand the harmony of nature. Fractals can also be found in other natural processes described by time-series measurements (i.e., noises, pitch and loudness variations in music, demographic data and others). Analysis of musical structure has revealed evidence of both fractal aspects and self-similarity properties in instrument tones and music genres. Voss and Clark [3] investigated aspects in music and speech by estimating the power spectra for slowly varying quantities, such as loudness and frequency. The fractal and multifractal aspects of different genres of music were analyzed by Bigrelle and Iost [4], where it was proposed that the use of fractal dimension measurements could benefit the discrimination of musical genres. Su and Wu [5] applied Hurst exponent and Fourier analysis in sequences of musical notes and noted that music shares similar fractal properties with the

This content is AI-processed based on ArXiv data.