Separation of Dominant Speech from Simultaneous Talkers

P.S. Sathidevi, S. Aravindakshan, and R. Rajavel (India)


CASA (Computational Auditory Scene Analysis), Cochlear filter banks (Gammatone, Lyons Passive Ear model, Discrete Wavelets), Pitch Estimation.


The hearing system, even in front of complex auditory scenes and in unfavorable conditions, is able to separate and recognize auditory events accurately. A great deal of effort has gone into the understanding of how, after having captured the acoustical data, the human auditory system processes them. For a hearing impaired person, it will be useful if we can separate out and amplify only the dominant speech from a mixture contaminated with babble noise (speech from other talkers). A computationally efficient algorithm for the separation of a dominant speech from a mixture of multiple speeches is proposed in this paper. For this, a wavelet packet filter bank matching to the critical bands of human cochlea is designed and implemented.

Important Links:

Go Back