Potential applications for automatic speech recognition

The Speechlab Processing team aims to be a premier R&D team to provide customized, relevant speech processing solutions. The programme was initiated to conduct R&D in speech recognition for Asian languages with poor language resource support, e.g. lack of good annotated speech corpus. The team also explores related speech technologies such as speech processing algorithms and keyword spotting for robust performance in various signal domains. The team’s current research capability includes speech enhancement, domain adaptation, conversational code-switch and far-field Large Vocabulary Conversational Speech Recognition (LVCSR).

Real time Code-switch Recognition system with recognition accuracy up to 85%
Speech indexing and retrieval with Malay Broadcast News and Singapore English Broadcast News and Conversational Speech

The team has achieved the following:

  1. Best Technology Awards from TL@NTU, Nanyang Technological University, Singapore, 2016 and 2018
  2. Ranked 3rd in the National Institute of Standards and Technology (NIST) 2018 Speaker Recognition Evaluation Challenge
  3. Best performance award for Oriental Language Recognition (OLR) Challenge 2017
  4. Ranked 3rd on both EER and minimum detection cost in NIST Speaker Recognition Evaluation Challenges, 2016
  5. No. 1 performance in Query-by-Example Spoken Term Detection (QbE-STD), 2015
  6. Ranked 3rd in NIST OpenKWS Evaluation, 2015

An example of our source separation system that separate mixture input speech into two separate speech. The system capability can be extended to more than two speakers
Assoc Prof Chng Eng Siong