Logo: University of Southern California

Research making noisy audio cleaner wins best paper award

Viterbi PhD students’ effort wins best paper award at the top conference, Interspeech 2013

October 03, 2013 —
EE PhD students Colin Vaz (center) and Vikram Ramanarayanan (right) with Professor Shri Narayanan (left)

Colin Vaz and Vikram Ramanarayanan, EE PhD students from the Signal Analysis and Interpretation Laboratory (SAIL) of the Ming Hsieh Department of Electrical Engineering have won a 2013 INTERSPEECH Best Paper Award. The award winning paper, co-authored with Professor Shri Narayanan, is entitled “A two-step technique for MRI audio enhancement using dictionary learning and wavelet packet analysis”.
The paper was presented at the 14th Annual Conference of the International Speech Communication Association held in Lyon, France in August 2013. Interspeech is the largest international meeting on speech science and technology.

The paper tackles the important problem of improving the quality of speech recorded in extremely noisy conditions in a two-step approach. First it uses a probabilistic latent component analysis to learn dictionaries of the noise and noisy speech portions to separate speech from noise. Then it applies a wavelet packet analysis based speech de-noising. The performance was demonstrated to be superior to the state of the art through objective and subjective human listening metrics. One target application for acquiring high quality speech in an MRI machine is to enable not only new speech production research but to make spoken communication with patients during scans possible.

Details of this research, and on going interdisciplinary efforts on speech signal processing and human speech production can be found at http://sail.usc.edu