SCS Undergraduate Thesis Topics

Fang Qiao Roni Rosenfeld Layperson-trained Speech Recognition for Resource Scarce Languages

We develop practical methods in speech recognition for low-resource languages overlooked by most effort in the field today. The resulting technique will potentially provide speech technology for languages of the developing world where many of the speakers are illiterate, by allowing one to build low-cost speech recognizers with high accuracy over small vocabularies involving minimal audio data and human expertise for training. Specifically, we will design pronunciation-generating algorithms for new languages based on existing speech recognition engines for other languages through cross-language phoneme mapping. We will also develop methods that will eliminate the need for language expert-based training.

Close this window