Speaker identification and verification from audio coded speech in matched and mismatched conditions

We investigate the effect of audio coding on speaker identification and verification when training and testing conditions are matched and mismatched. Experiments use popular audio coding algorithms (Windows Media Audio 9.1, Advanced Audio Coding, MPEG Audio Layer III) and a speaker identification and verification system based on Gaussian mixture models. There is some loss in identification and verification performance for audio coding process without the change of sample rate, and a great loss when sample rate changes during audio coding process.