Automatic Non-Native Speaker Recognition Systems

The objective of Phase II is to develop, demonstrate, and validate a novel implementation of a non-native language and speaker recognition system. In addition to the binary classification of native and nonnative speakers developed in Phase I, we will implement and evaluate several new methods to recognize possible native languages or dialects of nonnative speakers (a minimum of 5 languages). Recognition of non-native speakers is a multi-class recognition problem. Each accent or dialect will be considered as one class. We will collect a database of multiple dialects of a selected language, and then conduct research on the database. Our research will cover all existing approaches plus new algorithms developed in Phase I and II. The proposed system will be tested in both clean and noisy environments and its superior results will be validated by plugging it into a downstream speech recognition or machine language translation system. By the end of Phase II, we will provide the Army a working system to test. We will also provide detailed test plans, procedures, results and design documentation. Furthermore, we will deliver a matured non-native speaker recognition system, which can be used alone or in combination with speech recognition systems, machine translation systems, etc.