Employing Phonetic Speech Recognition for Language and Dialect Specific Search

Abstract

We discuss the notion of language and dialect-specific search in the context of audio indexing. A system is described where users can find dialect or language-specific pronunciations of Afghan placenames in Dari and Pashto. We explore the efficacy of a phonetic speech recognition system employed in this task.

Computational Approaches to Exploring Persian-Accented English

“…While such an approach shows promise, it suffers from the need to develop a potentially infinite number of classifiers depending on the particular L1/L2 pronunciation mismatch being explored. Miller et al (2014) employed the Nexidia phonetic speech recognizer (PSR) (Gavalda & Schlueter, 2010) to perform language and dialect specific audio search of Afghan toponyms in Dari (the Afghan variety of Persian) and Pashto. Figure 2 shows how PSR works.…”

“…During a searching phase, the user can specify a phonetic string and the system will return sequences of timepoints by which the phonetic lattice can be traversed through the sounds of that string above a user-specified level of confidence. Miller et al (2014) used such a system to search for the Dari and Pashto pronunciations of particular toponyms using both Dari and Pashto acoustic models. Such searches depended on crucial differences in pronunciation of particular speech sounds between Dari and Pashto and their dialects.…”

“…We outline an approach to automatic detection of pronunciation errors using an off-the-shelf speech recognizer based on the principles of Miller, Strong, Jones, and Vinson (2014). We then describe typical English pronunciation errors by speakers of Persian and show how they can be detected using our approach.…”