SMIIP Lab wins 3 prizes at Interspeech 2019 | Duke Kunshan University

SMIIP Lab wins 3 prizes at Interspeech 2019

October 31, 2019

Ming Li accepts an award on behalf of the SMIIP Lab at Interspeech 2019

Researchers from Duke Kunshan won three awards at the recent Interspeech 2019, the world‘s largest and most comprehensive conference on the science and technology of spoken language processing, in Graz, Austria.

The event, which attracted about 2,000 attendees, featured talks by leading figures in speech recognition and artificial intelligence (AI), as well as exhibitions, poster sessions and a series of high-tech challenges.

A team from Duke Kunshan’s Speech and Multimodal Intelligent Information Processing (SMIIP) Lab led by Ming Li, associate professor of electrical and computer engineering, received first prize in the computational paralinguistic challenge.

The lab also picked up second prize in the DIHARD speech diarization challenge and third prize in the VoxSRC speaker recognition challenge.

When humans talk, we send out a speech signal that not only contains lexical information but also delivers various paralinguistic attributes such as pitch, volume and intonation. These modify meaning, add nuance or convey the emotional state of the speaker.

The computational paralinguistic challenge, now in its 10th year, tests teams on their AI program’s ability to recognize paralinguistics including emotion, crying, voice pathology and dialect.

For this event, the SMIIP Lab researchers teamed up with experts from the Lenovo Research Institute to compete in two sub-challenges, in which they beat rivals from Sorbonne University in France, the Idiap Research Institute in Switzerland, Carnegie Mellon University and Germany’s Munich University.

The DIHARD challenge requires teams to determine who spoke and when in a multi-speaker environment using audio recordings. Teams from the University of Southern California, the Russian Voice Technology Center and Spain’s University of Zaragoza also took part.

To succeed in the VoxSRC challenge, competitors need to identify a celebrity speaker using YouTube videos of varying sound quality, from professionally edited interviews to red-carpet chats and casual conversations with background noise. The SMIIP Lab’s team finished in third place behind the Czech Republic’s Brno University of Technology and Microsoft.

Interspeech, which ran this year from Sept. 15 to 19, is organized annually by the International Speech Communication Association, a nonprofit organization based in France.

Li, who joined Duke Kunshan among the first faculty cohort for the undergraduate program in 2018, served as area chair of speech and language recognition at Interspeech 2016 and Interspeech 2018.