Low-resource automatic speech recognition and error analyses of oral cancer speech