Gang Liu

The National Institute of Standards and Technology (NIST) is coordinating a special speaker recognition challenge. This challenge is intended to foster interest in this field from the machine learning community and its early results will be featured in a special session of the Odyssey 2014 International Workshop on Speaker and Language Recognition. It is based on the i-vector paradigm widely used by state-of-the-art speaker recognition systems. By providing such i-vectors directly and not… Show more

The National Institute of Standards and Technology (NIST) is coordinating a special speaker recognition challenge. This challenge is intended to foster interest in this field from the machine learning community and its early results will be featured in a special session of the Odyssey 2014 International Workshop on Speaker and Language Recognition. It is based on the i-vector paradigm widely used by state-of-the-art speaker recognition systems. By providing such i-vectors directly and not utilizing audio data, the evaluation is intended to be readily accessible to participants from outside the audio processing field. The challenge is designed to foster research progress, including goals of:
Exploring new ideas in machine learning for use in speaker recognition
Making the speaker recognition field accessible to more participants from the machine learning community
Improving the performance of speaker recognition technology Show less

1 Addressing closed-set Language ID on large-scale and challenging dataset such as NIST LRE and DAPRA RATS, improve classification performance based on phonotactiv system (PPRLM & PRSVM) and acoustic system (GMM-UBM, state-of-the-art i-Vector system). Other involved techniques: voice activity detection, noise robust and discriminative feature extraction, effective back-end classifiers (PLDA, SVM, DNN, etc.) ,feature and back-end fusion, score calibration and normalization.
2 For open-set… Show more

1 Addressing closed-set Language ID on large-scale and challenging dataset such as NIST LRE and DAPRA RATS, improve classification performance based on phonotactiv system (PPRLM & PRSVM) and acoustic system (GMM-UBM, state-of-the-art i-Vector system). Other involved techniques: voice activity detection, noise robust and discriminative feature extraction, effective back-end classifiers (PLDA, SVM, DNN, etc.) ,feature and back-end fusion, score calibration and normalization.
2 For open-set Language identification, develop an efficient algorithm on munimum language selection (i.e. 85% reduction in required out-of-set language training data) to optimize out-of-set language rejection and overall classification performance based on the comprehensive distance measurement among world languages.
Show less

As one of the participating sites in the NIST Speaker Recognition Evaluation, the team was responsible for evaluating the state-of-the-art speaker recognition systems of their research group using NIST SRE 2012 data. The project involved leadership, team work, software development and various organizational skills.

This was the team’s first participation in the NIST Language Recognition Evaluation. They evaluated their front-end processing solutions in the context of a state-of-the-art language recognition system using NIST LRE 2011 data. The project involved leadership, team work, software development and various organizational skills.

As one of the participating sites in the NIST Speaker Recognition Evaluation, the team was responsible for evaluating the state-of-the-art speaker recognition systems using NIST SRE 2010 data. The project involved leadership, team work, software development and various organizational skills. Gang Liu was in charge of the front-end feature extraction.