Principal Investigator:

Ananth Sankar

Click here for other ivestigators



graphic_showing_idea Hierarchical Consistency Modeling is a DARPA-funded project in the Speech Technology and Research Laboratory at SRI International. This project aims to overcome improper statistical independence assumptions of traditional speech recognition algorithms, thereby forcing hypotheses formed during speech recognition search to be more self-consistent, and therefore more accurate. For example, in the picture, an adult female speaker is unlikely to switch to the voice of a child or a male in mid-sentence. Consistency modeling seeks to incorporate such fundamental aspects of human communication in algorithms and move significantly beyond accuracy levels afforded by present approaches. Specific areas of research include acoustic modeling, speaker and acoustic adaptation and normalization, noise-robust speech recognition.

  • Objective
  • Approach
  • Accomplishments
  • Future Plans
  • Project Publications