High Impact Factor : 4.396 icon | Submit Manuscript Online icon |

Singing Voice Separation Techniques Based on Spectral and Temporal Features: A Brief Review

Author(s):

Prashant Pralhad Zirmite , D.K.T.E. SOCIETY'S TEXTILE & ENGINEERING INSTITUTE, ICHALKARANJI; V. S. Kumbhar, D.K.T.E. SOCIETY'S TEXTILE & ENGINEERING INSTITUTE, ICHALKARANJI; S. D. Gokhale, D.K.T.E. SOCIETY'S TEXTILE & ENGINEERING INSTITUTE, ICHALKARANJI; V. B. Kumbhar, D.K.T.E. SOCIETY'S TEXTILE & ENGINEERING INSTITUTE, ICHALKARANJI; S. P. Salgar, D.K.T.E. SOCIETY'S TEXTILE & ENGINEERING INSTITUTE, ICHALKARANJI

Keywords:

MFCC, ZCR, PITCH, GMM, HMM

Abstract

To separate singing voice from music accompaniment the original file is segmented and each segment is analyzed for its voice content. Different features such as Mel-frequency Cepstral Coefficients (MFCCs), Zero Crossing Rate (ZCR), Log Frequency Power Coefficients (LFPC), Linear Prediction Coefficients (LPC), Pitch, timbre etc., are used to detect voice in segments. Again statistical classifiers as Support Vector Machine (SVM), Gaussian Mixture Model (GMM), Hidden Markov Model (HMM) etc. can be used to classify segments as voiced segment to separate singing voice. This paper represents some of the techniques for singing voice separation. According to their performance, specific technique can be selected for application.

Other Details

Paper ID: IJSRDV6I20685
Published in: Volume : 6, Issue : 2
Publication Date: 01/05/2018
Page(s): 1163-1166

Article Preview

Download Article