Raga Identification by using Swara Intonation
In this paper we investigate information pertaining to the intonation of swaras (scale-degrees) in Hindustani Classical Music for automatically identifying ragas. We briefly explain why raga identification is an interesting problem and the various attributes that characterize a raga. We look at two approaches by other authors that exploit some of these characteristics. Then we review musicological studies that mention intonation variability of swaras across ragas, providing us a basis for using swara intonation information in raga recognition. We describe an experiment that compares the intonation characteristics for distinct ragas with the same set of swaras. Features derived from swara intonation are used in a statistical classification framework to classify audio segments corresponding to different ragas with the same swaras.
Raga Malkauns Revisited with Special Emphasis on Modeling
A great deal of attention has been paid in the recent past on modeling musical structure and performance. In the present work, we model the structure of raga Malkauns using statistics.
Effect of Tempo on the Frequency Ratio of Notes in Singers and Instruments
Carnatic music consists of singing seven notes in three tempos (slow, medium, fast). It is important that a singer is able to maintain the frequency of each note at all tempos. But whether the human vocal mechanism is able to match the accuracy of a musical instrument with adequate training is a matter of curiosity. Hence, the present study considered two groups of trained and untrained singers and one group of Veena players. They were instructed to produce the ascending and descending notes of Ma:ya:ma:lavagowla ra:ga: in three tempos. Frequency of each note was measured and frequency ratio between the base note /sa/ and other notes were calculated and compared with those of the Veena. Results revealed that with increase in tempo, even the trained singers were not able to match the instrument in terms of the frequency ratio of notes.
Automatic Carnatic Raga Classificaton
Raga is the central melodic concept of two distinct systems of music in India, Hindustani (North Indian) and Carnatic (South Indian) music. Previous work demonstrated that pitch-class distributions (PCDs) and pitch-class dyad distributions (PCDDs) could be effectively used for raga classifcation in North Indian classical music (NICM).We find that PCDs and PCDDs are also effective for Carnatic raga classification, despite substantial differences in the manner of presentation, ornamentation, and melodic types as compared with NICM. In a thirty target experiment a 92.4% classification accuracy was achieved using a Bayesian classifier with both PCDs and PCDDs together. This ability of PCD and PCDD features to generalize to a novel musical context suggests that they are effective for capturing essential melodic characteristics of Indian classical music.
On the Possibility of Objective Assessment of Students of Hindustani Music
A tentative approach for quantitative assessment of progress of students of Indian classical music is proposed. In the present study only the accuracy of notes as well as their variations and the understanding of the general concept of a raga are attempted. However, for a fuller assessment voice quality, pronunciation, improvisation, range, rhythm, use of ornamentation etc. need to be considered. The basic physical parameter used for the present purpose is the pitch profile of the songs.
The procedure consists of extracting pitch profile and identification of notes. From these, four parameters are used for analysis. The inherent problem associated with an objective assessment in the Indian scenario is the absence standards regarding the frequency position of the notes as well as a unanimous position about a clear concept of a raga. It is therefore decided to use the raga song by a maestro whom the student considers as his idol for the standard template for comparison. The data consists of pitch sequences extracted from the songs of the students and the maestros. The ragas used for analysis are nine in number. The number of students is seven.
The analysis shows reasonable compliance of most of the students with their chosen maestros. The relevance of each parameter reflected through the analysis of the data is also presented.
Kalman Filtering Application to Corrupted Vocal Recording
In this paper, the Kalman Filter has been applied to clean a segment of a song sung by Rabindranath Tagore with musical accompaniments as available from an archival recording. It is observed that significant noise removal is achieved as evident audibly and supported by the spectrogram provided the order of the Autoregressive (AR) model used to represent this real system is high (p=30 or more) compared to the typical model order (p=10 to 15) as used for speech only. A quantitative measure of the noise removed has been defined in this paper as the Noise Removal factor (NRF). It is observed that the NRF reflects the observations as evident from the spectrogram.