Crowdsourcing the Pronunciation of Out-of-Vocabulary Words

Sajad Shirali-Shahreza, Pieter Luitjens, Natalie Morcos, Wen Xiao, Zhenghong Qian, Gerald Penn

Abstract: We propose a method for crowdsourcing the pronunciation of out-of-vocabulary words, which in our experiments has generated a lexicon of competitive quality to the CMU pronunciation dictionary. In contrast to an earlier approach, we use crowdsource workers to generate new pronunciations which are phonetically transcribed by an acoustic model, rather than merely to select among candidate alternatives from a letter-to-sound algorithm.