A Simple Method For Melodic Classification Based On Scale Analysis

Download as pdf or txt
Download as pdf or txt
You are on page 1of 2

A SIMPLE METHOD FOR MELODIC CLASSIFICATION BASED ON

SCALE ANALYSIS

Anas Ghrab
Higher School for Music - University of Sousse (Tunisia)
[email protected]

1. INTRODUCTION 3. GENERAL METHODOLOGY

One objective of the analysis of modal music is the com- This method consists of an extraction of the fundamental
parison of melodies, which implies their classification. While frequencies of each melody, where we use the Predominant
conventional methods of analysis are used to enter a modal Melody Extraction algorithm (Salamon & Gómez, 2012)
specificities bounded together, and the number is typically which is relatively robust for the detection of the melody
reduced, it is clear that the pertinence of the results is closely in a complex context.
related to the amount of data analyzed. We propose a gen- A Kernel Density Estimation is then applied to the fre-
eral method, implemented in Python, for the analysis of quency list to obtain Probability Density Functions (PDF-
the scale of melodies and to measure the melodic nearby. s) related to dominant frequencies in the melody. The peaks
We apply it to the analysis of a segmented file and several of the PDF gives as the frequencies that represent the scale,
songs of women from different regions of Tunisia. which are faced to most known intervals in an epimoric
(n+1/n) definition. For that we need first to detect the fre-
For the purpose of this analysis, we have implemented
quency that could represent the tonic note. We detail below
different functions in the python module Diastema 1 .
our approch for that.
When comparing melodies from different files, we have
to transpose them on a single reference frequency. Then we
2. BACKGROUND classify Probability Density Functions by the linear Corre-
lation Coefficient : the classification of PDFs should match
Several works the MIR field have focused on melodic anal- a melodic nearby.
ysis. Part of these works deal with melodies using a sym-
bolic representation (De León et al., 2004; León & Iñesta,
4. THE DIASTEMA TOOL AND EXAMPLES
2004; Li & Sleep, 2004; Frieler & Müllensiefen, 2005).
Other studies, such those of Bars Bozkurt (Bozkurt, Bozkurt) 4.1 First example
focused mainly on makamique analysis. As the special fea-
Our first example is the analysis of Violin Taksim.
ture of maqm is in its intervallic system, which is not taken
into consideration by the symbolic data, Bozkurt use dif- 4.1.1 PDF-s of the different segments
ferent basic detection algorithms to obtain representative
frequencies of the melody. These frequencies, retuned into 4.1.2 The detection of the tonic frequency-note
the same octave and converted using Holderian Comma, Useing different percentages of the last frequencies :
allowed the representation of histograms of main notes of
melodies (template-matching). Ioannidis et al. (2011) present [1.5%,2%,5%,10%,15%]
an extension to work Bozkurt using HPCP. The willingness P1 : [318, 318, 318, 318, 318]
of these approaches is to determine the scales of the main P2 : [318, 318, 318, 318, 347]
modes of Turkish music or to identify the maqm of a mu- P3 : [318, 343, 318, 318, 318]
sic compared to a predetermined scale (theoretical tuning P4 : [316, 316, 316, 318, 318]
systems). We can consider the we are here in a supervised P5 : [318, 318, 318, 318, 318]
approach.
The method presented here is different. Our main goal 4.1.3 The global PDF and the scale
is to analyze and classify melodies in an unsupervised man- [[’0.00’, ’1/1’, ’+’, ’0.00’],
ner. This approach should allow us to address and rede- [’36.54’, ’12/11’, ’-’, ’1.25’],
fine makam-s and modal scales of different art music in [’88.39’, ’9/8*12/11’, ’-’, ’0.56’],
a broader context. In addition, it is applicable to non- [’114.26’, ’4/3’, ’-’, ’10.68’],
scholarly traditional music, where melodies do not belong [’146.52’, ’4/3’, ’+’, ’21.58’],
to a pre-established modal theory. [’181.95’, ’3/2’, ’+’, ’5.86’],
[’209.70’, ’3/2*10/9’, ’-’, ’12.15’],
1 https://github.com/AnasGhrab/diastema [’258.89’, ’3/2*6/5’, ’+’, ’3.62’]]
Figure 1: PDF-s of the different segments; Global PDF from all segments

4.2 Second example Ioannidis, L., Gómez, E., & Herrera, P. (2011). Tonal-based
retrieval of Arabic and Middle-East music by automatic
303 audio files of traditionnal women songs from differ- makam description.
ent regions of Tunisia. They are field recordings made be-
tween 2007 and 2015 2 . León, P. J. P. D. & Iñesta, J. M. (2004). Statistical descrip-
tion models for melody analysis and characterization. In
In Proceedings of the 2004 International Computer Music
Conference, (pp. 149–156).

Li, M. & Sleep, R. (2004). Melody classification using a simi-


larity metric based on kolmogorov complexity. Sound and
Music Computing, 126–129.

Salamon, J. & Gómez, E. (2012). Melody extraction from poly-


phonic music signals using pitch contour characteristics.
Audio, Speech, and Language Processing, IEEE Transac-
tions on, 20(6), 1759–1770.

Figure 2: Self-similarity matrix

5. CONCLUSION
This paper presents a musicologial point of view. Its ap-
proach has to be extended with specialists in data-mining
and programmers.

6. REFERENCES
Bozkurt, B. An automatic pitch analysis method for turkish
maqam music. 37(1), 1–13.

De León, P. J. P., Pérez-Sancho, C., & Inesta, J. M. (2004).


A shallow description framework for musical style recog-
nition. In Structural, Syntactic, and Statistical Pattern
Recognition (pp. 876–884). Springer.

Frieler, K. & Müllensiefen, D. (2005). The simile algorithm


for melodic similarity. Proceedings of the Annual Music
Information Retrieval Evaluation exchange.
2 See the collection “Chants de femme dans les regions
tunisiennes” at the catalogue of the National Sound Archives
(http://phonotheque.cmam.tn), hosted by the Centre of Arab and
Mediterranean Music (http://www.cmam.tn).

You might also like