A framework for creation of telephone, cellular and VoIP speech corpus
Suparnakanti Das, Anupam Mandal, K. R. Prasanna Kumar, Paromita Choudhury, Anil Kumar Chilli · 2013
State of the art techniques used to automate speech processing for different applications are data-driven. The statistical methods derive robust models from large speech data characterizing variations of speech, channels and background. The increased convergence of multi-channel voice communication only adds to the need of representative data for consistent performance. There have been concentrated efforts towards collection of data to address such needs. The available technical literatures on speech corpus creation do not describe the framework of the collection infrastructure for such requirements. The paper presents an approach to design of integrated framework for collection of multi-channel voice communications over telephone, cellular and VoIP. The framework accommodates different requirements of corpus collection for different speech processing tasks. A realization of the design is illustrated using open IP-PBX asterisk.