Phrase-based rĀga recognition using vector space modeling

Abstract

Automatic rāga recognition is one of the fundamental computational tasks in Indian art music. Motivated by the way seasoned listeners identify rāgas, we propose a rāga recognition approach based on melodic phrases. Firstly, we extract melodic patterns from a collection of audio recordings in an unsupervised way. Next, we group similar patterns by exploiting complex networks concepts and techniques. Drawing an analogy to topic modeling in text classification, we then represent audio recordings using a vector space model. Finally, we employ a number of classification strategies to build a predictive model for rāga recognition. To evaluate our approach, we compile a music collection of over 124 hours, comprising 480 recordings and 40 rāgas. We obtain 70% accuracy with the full 40-rāga collection, and up to 92% accuracy with its 10-rāga subset. We show that phrase-based rāga recognition is a successful strategy, on par with the state of the art, and sometimes outperforms it. A by-product of our approach, which arguably is as important as the task of rāga recognition, is the identification of rāga-phrases. These phrases can be used as a dictionary of semanticallymeaningful melodic units for several computational tasks in Indian art music.

ragamAI: A Network Based Recommender System to Arrange a Indian Classical Music Concert

“…Machine learning has been used in the Indian classical music over the past decade to study multitude of concepts. For example, classifying recordings from YouTube based on the swara using random forest algorithm [13], using pitch information in music signals [14], and a vector based classification model [15], similar to text classification model. Identifying tonic of a music from multi-pitch analysis of the given audio was also given as a classification problem [16].…”

Morty

“…Recently, MORTY was used as a benchmark for raga/raag recognition of audio recordings of Hindustani and Carnatic music in comparison with two novel methods [15,16]. Below we explain the results briefly.…”

“…In many music cultures there is no standard reference tuning frequency, which makes it crucial to identify the tonic frequency to study melodic interactions. Estimating the tonic of a recording is the first step for various computational tasks such as tuning analysis [7], automatic transcription [4] and melodic motif discovery [16].…”

“…There has been a extensive interest on mode recognition in the last decade [17]. Most of these work focus on culturespecific approaches for music traditions like Ottoman-Turkish Makam music (OTMM) [13], Carnatic music [11,12,16], Hindustani music [9,10,15] and Dastgah music [1]. A considerable portion of these studies are based on comparing pitch distributions [9,10,11,12,13], which are shown to be reliable in the mode recognition task.…”

“…A considerable portion of these studies are based on comparing pitch distributions [9,10,11,12,13], which are shown to be reliable in the mode recognition task. There also exists recent approaches that are based on characteristic melodic motif mining using network analysis [15,16], aggregating note models using automatic transcription [18] or audio-score alignment [23] and classification using neural networks [24,26], all of which are designed specific to the studied music culture and are not generalizable to other music cultures without considerable effort. Similarly, several studies on tonic identification use pitch distribution based methods [6,10].…”