Forensic speech scientists may sometimes be faced with the task of extracting information about an unknown speaker in a recording. It is proposed here that accent recognition technology could assist analysts in such cases and we begin to explore the Y-ACCDIST system’s potential for this purpose. Research on Y-ACCDIST so far has largely focussed on its ability to distinguish between varieties which are much more similar to one another than previous automatic accent recognition research [1].
The experiments presented here build on this and challenge YACCDIST in other ways relevant to forensic applications: spontaneous speech data and degraded data