Call For Paper Volume:4 Issue:8 Aug'2017 |

Speech/Music Differentiation and Male/Female Voice Diagnosis in Speech: A Hierarchical Approach

Publication Date : 01/07/2015



Author(s) :

Arijit Ghosal , Suchibrota Dutta , Debanjan Banerjee.


Volume/Issue :
Volume 2
,
Issue 6
(07 - 2015)



Abstract :

In this work, we have presented a simple hierarchical scheme for differentiating audio signals first into speech and music and further speech into male speech and female speech. In the first level, audio signal is classified into two sub-classes speech and music (music includes both instrumental and song) based on some popular salient low level time-domain acoustic features which are very closely related to the physical properties of source audio signal. Zero crossing rate (ZCR), short term energy (STE) and “delta-energy”, which is the very small change of energy of an audio signal, are used as feature in the first level of classification. The strength of the feature-set is further enhanced by incorporating features computed from the co-occurrence matrix of delta-energy. For the second phase of classification some popular low level frequency-domain acoustic features are used as male voice differs with female voice mostly in frequency domain. Along with low level frequency-domain features, low-level time-domain features are also used in male/female voice diagnosis in speech. We have used low level frequency-domain features based on spectral flux and spectral centroid and low level time-domain features based on ZCR and STE for this purpose. RANSAC and Neural-Net has been used at all stages as classifier. The experimental result denotes the efficiency of the proposed scheme.


No. of Downloads :

11


Indexing

Web Design MymensinghPremium WordPress ThemesWeb Development

Speech/Music Differentiation and Male/Female Voice Diagnosis in Speech: A Hierarchical Approach

June 30, 2015