North Indian Classical Music's Singer Identification by Timbre Recognition using MIR Toolbox

Abstract

Timbre recognition has always been associated with musical instrument identification. In music information retrieval (MIR) community there has always been increasing interest in identifying singers and identifying the musical instruments. There can be unending debate on, whether to consider human throat as a kind of musical instrument generating sound through air pressure and vibration or not. Any musical sound possesses "Timbre", an unidentified and undefined entity which is yet non-tangible that uniquely defines the sound. A lot of research has been done on catching this fuzzy term in terms of its synonyms related to light or texture or so. MIR toolbox of MatLab provides some strong techniques to extract variety of audio attributes/characteristics from an audio file. These attributes are called as "Audio Descriptors". In this paper, we have studied such audio descriptors that fall under the Timbre category and found out the most prominent audio descriptors that plays substantial role in the identification of a singer from North Indian classical music. In this system we have considered Noisy data, Noise filtered data without background instruments and noise free studio recordings with only Tanpura instrument in the background as input. We have found that roll off, brightness, roughness and irregularity are the four strong audio descriptors designated under the Timbre category of MIR Toolbox, plays vital role in the identification of a singer from North Indian Classical Music giving accuracy of identification of 96.66% for three singers trained and tested simultaneously on studio recordings containing Tanpura(a Supportive Musical instrument of the singer) each of 5 sec duration and sampling rate of 11,025Hz, 16 bits with Pulse Code Modulation (PCM) uncompressed file format. The results indicate that a singer can also be treated as a wind type of musical instrument and be successfully identified by recognizing its Timbre.

Robust singer identification of Indian playback singers

“…Carnatic singer identification was proposed for six singers by Sridhar and Geetha [29] for efficient music retrieval of Carnaticbased songs. Saurabh et al [30] identified ten singers from North Indian classical music using timbre descriptors for noise-free studio recordings accompanied with continuous background music of tanpura and violin/harmonium/flute, etc., and achieved an accuracy of 58.33%. A technique of identifying singers using video songs from the Internet and CD for Indian playback singers using cepstral coefficients is presented in [31,32], however tested for a lower corpus.…”

On the Study of Thai Music Emotion Recognition Based on Western Music Model

“…Therefore, the development of a music emotion analysis system is a challenge for researchers today, especially research on non-western music since many research works 2 are still based on western music. We found a few non-western works such as Sinhala (Sri Lanka) songs [2] and Indian (India) songs [3].…”

Singer Identification Using Statistical Signal Processing

“…It is used to determine whether a random variable comes from a population which is normally distributed (Shenoy et al, 1994). For experiment purpose 5 Musical features or Audio features like spectral centroid, entropy, skewness, kurtosis and brightness are used as a dataset (mahto et al, 2014;Datta et al, 2017;Chandwadkar et al, 2013;Deshmukh et al, 2014;Masood et al, 2016). Normality test or (Kolmogorov-Smirnov test) is proposed on spectral centroid, entropy, skewness, kurtosis and brightness, concerning a singer's voice.…”