SLS: Low Resource Multi-lingual Speech Recognition
Conventional speech recognizers require hundreds or thousands of hours of annotated data for training. As a result, speech recognition capability is only available for about 100 of the worlds 7000 languages. In this research, we are developing speech recognition for languages that have fewer linguistic resources available, such as Cantonese, Pashto, Turkish, and Tagalog. The immediate goal of this project is to help us develop tools for rapid deployment of speech recognition capability for new languages. We will also investigate language properties and how they can be used to improve performance on other languages. We are looking for someone with scripting language and Perl/Python expertise. Experience with machine learning is desirable, but not required. Knowledge of any particular language is also not required. If interested, contact Ekapol Chuangsuwanich (email@example.com) for further information. Possibility to turn this project into a Super UROP.