Recent trends and challenges in speech-separation systems research — A tutorial review

The pioneering work on the `separation of speech from mixture of acoustic sources' dates back to as early as 70s and since then, two main approaches namely traditional approach using signal-processing techniques and computational auditory scene analysis (CASA) approach using auditory-modeling methods have been concurrently attempted by researchers to find solution to the problem of what is known as `cocktail party' effect. This field has gained momentum in the last decade, as interest has been constantly growing among the researchers in this field. However, the field itself is still in its infancy and immature, as we are yet to see a device or gadget that implements speech separation algorithm as a commercial product for use by the community. The main reason for this is the lack of clear understanding of the processes and mechanisms involved in human auditory system for developing the theory and models for segregation of sounds. This paper is not intended to be an exhaustive review of speech-separation systems development. Rather we focus on a number of key issues and their historic relationship. The paper projects this research topic to assist novice researchers in this field and experienced researchers to re-focus their directions. This paper concludes by projecting the major issues and challenges facing the speech research community in realizing systems based on speech separation for the future and suggests new directions for the expedition of the growth of this important fascinating field.