wSPIRE: A Parallel Multi-Device Corpus in Neutral and Whispered Speech

Download PDF
Open Webpage

Bhavuk Singhal, Abinay Reddy Naini, Prasanta Kumar Ghosh

Published: 01 Jan 2021, Last Modified: 03 Oct 2025O-COCOSDA 2021EveryoneRevisionsBibTeXCC BY-SA 4.0
Abstract: Most of the speech technologies for whispered speech are lagging behind due to the scarcity of data. Hence, in this paper, we present and open source a multi-device parallel speech corpus in neutral and whispered mode, called wSPIRE. The wSPIRE consists of 88 (54 males and 34 females) speakers recorded using five recording devices in neutral and whispered modes. So this dataset contains almost 44000 audio recordings resulting in a total of ~36 hours speech corpus. Apart from this, wSPIRE corpus also contains word-level annotations for each recording. Different types of analyses on the speech recordings and word level annotations are presented. We also throw light on the possible research areas and problems which can be addressed using the wSPIRE.