From: Robust singer identification of Indian playback singers
Acoustic feature | Experimental database | Song duration | Song level accuracy | Remarks |
---|---|---|---|---|
Formant + HSE + Vibrato + timbre | Bollywood singers 26 singers (10F, 16 M) 550 songs | 3–9 s | 86% | Proposed work |
Method I: Carnatic interval cepstral coefficients Method II: MFCC [29] | Carnatic singers 6 singers 116 songs | – | Method I: 87% Method II: 56% | Lower singer dataset |
Timbre [30] | North Indian singers 10 singers | 5 s | 58.33% | Lower singer dataset |
Timbre + pitch + MFCC + LPC [31] | 6 singers (2F, 4 M) 300 songs | 180 s | 81% | Audio portion taken from video songs. Lower singer dataset |
MFCC + spectral features [32] | Tamil singers 3 singers (2F, 1 M) 300 video songs | 10 s | 95% | Audio portion taken from video songs. Lower singer dataset |
Cepstral mean subtracted MFCC [34] | Bollywood singers 20 singers (6F, 14 M) 500 songs | 60–166 s | 84.5% | Longer Song duration |
Perceptual linear prediction [50] | Tamil singers 10 singers (5F, 5 M) 200 songs | 20 s | 55.56% | Lower singer dataset |