| This research investigates the use of Automated Speech Recognition (ASR) to predict Human Speech Recognition (HSR) in communication systems. An ASR algorithm, derived from bivariate autoregressive models of binaural recordings, and an index, based on the physical interpretation of the ASR system in terms of information channel capacity, are proposed. This algorithm and index uses recordings from an artificial head to simulate a Modified Rhyme Test (MRT), which evaluates the ability of human subjects to recognize speech correctly or incorrectly when mixed with different levels of noise. This speech intelligibility tool is verified for human jury testing conducted for a variety of environments and types of noise. In addition to quantifying the limitations of this index, its applicability to hearing-impaired subjects and other types of speech intelligibility tests are explored. This index can eventually lead to implicit cost savings by reducing the need for human jury testing and by providing a less-subjective approach to evaluation of communication systems. |
