Ragas demonstrate the pride of Indian classical music. Raga is the original musical form in Indian classical music. It consists of set of swaras (lyrical notes) that made up of various characteristics as a melodious conception which is played by the instruments and the singer. Based on the features of the raga, the Indian classical music is separated into two Parts: Hindustani (North Indian) and Carnatic (South Indian) classical music. Our experiment is concentrate on Hindustani classical music. In our experiment, K Nearest Neighbor (KNN) and Support vector machine (SVM) classifiers are used on the raga dataset of Yaman and Bhairavi to achieve classification and identification of the raga. We have done accurate outcomes with both KNN and SVM classifiers.
Indian classical music is the music of the Indian subcontinent. Raga or Raag hold a prominent position in Indian Classical Music. A Raag is a collection of musical notes that, when sung or performed on a musical instrument, are quite attractive. Raga recognition comprises of methods that define and classify notes from a piece of music into a suitable raga. In Hindustani classical music, Ragas is a very significant idea and express the moods and sentiments of concert. The classification of ragas comes only after an enough amount of exposure as it is an intellectual process.1 Any of the attributes of ragas have to be translated into appropriate characteristics for automated recognition. The characteristics of Ragas are based on Indian Classical Music techniques, which blend notes with the following features to qualify as a Raga.
There have to be at least 5 or 7 notes (swaras) in a Raag. The primary seven notes are S (Sa), R (Re or Ri), G (Ga), M (Ma), P (Pa), D (Dha), N (Ni).
Each Raga or Raag is composed of a "Aaroh" that implies swaras scale up and a "Avroh" that implies swaras scale down.
Each raag consisting of "Vadi" means main notes and "Samvadi" means supporting swaras.
It has a constant frequency rate. Notes in a raga are a series of continuous (back and forth movement in a rhythm) variation, such sort of notes are known as Gamakas.
A set of Swaras which are distinctively recognizes a raga. There is a particular Pakad for each raga.
Tala refers to a rhythmic form, which is constructed from variety of beats.
Thaat is used in raga classification. There are unique ten Thaats namely Kalyan, Bilawal, Bhairav, Khamaj, Poorvi, Marwa, Kafi, Asawari, Bhairavi, and Todi.
The piece of music has to be converted to Swara
for classification. Due to the following factors,
there are several difficulties in converting the
piece of music in Swara [
The key purpose of raga recognition is that it will provide a good start for Hindustani music information retrieval and it allows us to predict the raga's performance and accuracy. Besides this, for music analysis, we can also create a playlist focused on ragas.
In this section, we have reviewed different work
done by other authors and analyzed their work for
future scope. We have tried to give an analysis of
various classifiers, their relevance and
performance for Raga identification. In [
In this section, we will discuss different characteristics of audio and Machine Learning algorithms like K Nearest Neighbor and Support Vector Machine in a brief way.
Each audio signal comprises of several features. But, it requires fetching the characteristics that are suitable for the issue that we want to solve. The method of fetching characteristics to apply for the study is referred to as feature extraction. We will have a brief idea about some of the characteristics below, in detail.
A spectrogram is a graphical demonstration of the spectrum of frequencies of a signal as it differs with time. When it is used with an audio signal, spectrograms are sometimes referred to as the sonographs, voiceprints or voicegrams. To determine the raga, we are using the mean of spectrogram to get which tone/ pitch is used more.
This feature is one of the most necessary techniques to extract attributes of an audio signal and it is used mostly when we are working on audio signals. The mfccs of a signal are a set of characteristics (approximately (10–20)) which in brief illustrates the general form of a spectral cover.
It shows that the "center of mass" that considered the weighted mean of the frequencies present within the sound. If it gets the equal frequencies in tune for a particular time span, then the spectral centroid might be around the center and if there are excessive frequencies at the end of the sound then the centroid tends to be closer to its end.
The rate at which sign varies is known as the zerocrossing rate. Zero crossing rate is the rate wherein the signal varies from positive to negative and vice versa. Speech recognition and music information retrieval are being commonly used in Zero crossing rates. It has excessive values for loud and noisy sounds like in metal and rock.
Particularly Roll-off suggests the activity of a particular sort of channel; one planned to Roll-off frequencies raised or lowered at a certain point. It is called roll-off as the method is progressive.
A radiated spectral quantity is not less than half its maximum value in spectral bandwidth. It determines the extent of the Spectrum.
This is an interval difference between lower and higher frequency.
During our analysis, we have found that the Supervised Machine Learning approach might be a good fit for our problem.
We have tried to implement various classification algorithms and found that K Nearest Neighbor and Support Vector Machine is quite appropriate for our experiment.
K Nearest Neighbor (KNN) is a supervised learning method. It is the simplest but robust algorithm that is applied for both regression and classification problems. To build a prediction, the KNN algorithm uses the whole dataset in which we attempt to classify data points to a particular category with the help of the training set. Support Vector Machine (SVM) Support vector machine (SVM) is alsoa supervised learning method that is usually used for classification. A hyperplane that clearly divides the sampling points with various labels is identified by this algorithm. It separates sample points of both labels and class on different sides of the hyperplane.
In this experiment, we have used the dataset of Audio files. The dataset was created by extracting 60-second audio clips from the internet. For music and audio analysis, a python package Librosa is used. It provides the segments that are required for creating music information retrieval systems. Another open-source machine learning library is Scikit-learn which supports both, supervised and unsupervised learning methods. It also provides a variety of tools for model selection, data preprocessing, model fitting, and estimation. In this experiment, we have chosen Yaman and Bhairavi Raga. We have split them into 60-sec frames which allow the computer to work only on the specific part of the song like Pakad, Aaroh, and Avroh and remove other noise and empty fields from the audio. Using Librosa Library, we have created a .CSV file to save all the features like Mfcc, Spectrogram, Bandwidth, Centroid, zerocrossing, and Roll-off which are to be extracted from the audio file. Below is the flow of the process.
Audio Dataset
Feature Extraction
Identification and Classification This is the visual representation of the classification procedure in which features are taken out from the audio frame and compared with the weight of the closer mean. Using Scikit Learn library, we have implemented KNN and SVM algorithms on the .CSV data file. We found that both KNN and SVM fit best for the classification of raga other than Logistic Regression.
In our experiment, we have chosen Ragas like Yaman and Bhairavi. We have used vocalinstrument Dataset consisting of 341 audio clips, out of which, 194 audio clips are of Yaman and 147 audio clips are of Bhairavi.
The tables below show the accuracy of KNN and SVM algorithms.
KNN
1 2 3 4 Train/ Test ratio into 3 different categories that are 80/20, 60/40, and 40/60 respectively. For all KNN values, we got the highest accuracy in Train/Test ratio 80/20 i.e. For KNN value 1 to 4; we received 98%, 97%, 97%, and 95% accuracy respectively. Using SVM, we got 95% accuracy for the 80/20 Train/Test ratio.
From this comparison, we can conclude that KNN with all Neighbor values for Train/Test ratio 80/20 gives the highest accuracy which is better in comparison with SVM.
A short introduction on raga and its attributes are considered. Prior policies for raga classification and recognition are observed with their data records, implementation applications, correctness, and problems. In this paper, we have discussed the classification of different ragas like Yaman and Bhairavi by applying K-Nearest-Neighbor (KNN), Support vector machine (SVM) machine learning algorithms. We have obtained good results with KNN and SVM, but in our experiment, KNN seems to be performing slightly better.
In the future, we will expand our dataset with many other Ragas for acquiring more accurate results and also implement other classifiers for detecting ragas in a well-defined manner.