Accent versus Mispronunciation: modifying speech recognition software to recognize speech and not a perfect accent in the ESL context

In this presentation, we will detail our attempts to use speech recognition (SR) software in the ESL classroom for the purposes of interview and speaking assessment. Currently, the most effective SR technology depends on modifying its program to suit your voice within certain parameters such as vocabulary size and regional dialects. Other SR technology on smartphones and Internet apps, require you to modify your grammar or copy a standard accent/dialect. While these have been found to be useful, they are not appropriate for our needs. We require SR technology to serve as an automated practice tool for interviews and not serve as a pronunciation judge. Many Japanese speakers of English have an acceptable level of pronunciation, however their accent prevents them from using SR software. Our goal is to modify current SR technology, so that perfectly understandable Japanese-accented speakers can use our VIS for practice and improving their interview skills.