A linguistically-informative approach to dialect recognition using dialect-specific context-dependent phonetic models

In this work, we explore automatic approaches to learn dialect discriminating pronunciation patterns and use these patterns to automatically recognize dialects. Since linguistic literature suggests that dialect differences often occur in certain phonetic contexts [2, 7, 8, 9], we extend adapted phonetic models [Shen et al. (2008)] to consider phonetic contexts. We evaluate our system on classifying American and Indian English. Despite many challenges (e.g., subdialect issues and suboptimal phone recognition accuracy due to lack of word transcriptions), we discover dialect discriminating biphones compatible with the linguistic literature, while outperforming a baseline system by 7.5% (relative). Our work is an encouraging first step toward a linguistically informative dialect recognition system, with potential applications such as forensic phonetics and accent training tools. [This work is sponsored by the Command, Control and Interoperability Division (CID), which is housed within the Department of Homeland Security’s Science and Technology Directorate under Air Force Contract No. FA8721-05-C-0002. N. C. F. is also supported by the NIH Ruth L. Kirschstein National Research Award and NIH/NIDCD Grant Nos. DC02978 and T32DC00038.]