By

Text-dependent Speaker Verification and Identification (1 April 2018 – 31 March 2019)

Speaker verification aims to authenticate a person based on the voice example. Broadly it can be categorized as two categories, text-dependent and text-independent. From the view of application based systems text-dependent speaker verification is preferred due to involvement of short pass phrases during training and testing. This project aims to develop novel and effective methods for text-dependent speaker verification and identification to use in real-world applications.

Some of our explorations for text-dependent speaker verification include utterance compensation framework and unified framework for speaker and utterance verification. In the utterance compensation framework, a background utterance model is trained that is used to compensate the lexical content information as we are interested towards the speaker-specific information. Utterance verification is another that co-exists with text-dependent speaker verification. The general way of utterance verification is to have a separate framework from the speaker verification system. We propose a unified framework that can perform both speaker and utterance verification together. It is referred to as unified framework as speaker-utterance-verification. Further, we also aim to explore near-field vs. far-field and wake up word based speaker verification under the scope of this project.

Project Duration: 1 April 2018 – 31 March 2019

PUBLICATIONS

Conference Articles

  • Tianchi Liu, Rohan Kumar Das, Maulik Madhavi, Shengmei Shen and Haizhou Li, “Speaker-Utterance Dual Attention for Speaker and Utterance Verification”, in Proc. INTERSPEECH, Shanghai, China, October 2020.
  • Tianchi Liu, Maulik Madhavi, Rohan Kumar Das and Haizhou Li “A Unified Framework for Speaker and Utterance Verification” in Proc. Interspeech 2019, Graz, Austria, September 2019. [link]
  • Wei Rao, Chenglin Xu, Eng Siong Chng and Haizhou Li, “Target Speaker Extraction for Multi-Talker Speaker Verification”, in Proc. Interspeech, Graz, Austria, September 2019. [link]
  • Rohan Kumar Das, Maulik Madhavi and Haizhou Li “Compensating Utterance Information in Fixed Phrase Speaker Verification” in Proc. Asia-Pacific Signal and Information Processing Association (APSIPA) Annual Summit and Conference (ASC) 2018, Honolulu, Hawaii, USA, November 2018. [link]