Noise Estimation for Speech Enhancement

Add to your list(s) Download to your calendar using vCal

• Jionjun Bai
• Tuesday 02 February 2010, 16:00-17:00
• 503.

If you have a question about this talk, please contact Karen Lewis.

Imaging yourself moving to a new flat next to a railway line, at first you might be distracted and annoyed by the noise of trains passing by. Over time, you will start to get used to it, and barely notice its existence. Surprising by our own adaptability, we have been trained passively, and thus being able to learn and predict when the same type of noise occurring over and over again. It seems that our brain has a filter that are capable of removing these registered noise. Motivated by such mechanism, we have proposed a hidden Markov model (HMM) that is trained continuously based on the past noise spectrum, and being able to predict and estimate the noise signal in the future. The noise signal will be grouped into different states, with each state representing a distinct type of the noise. The results show a consistent improvement in noise estimation over a conventional moving average filter. A novel method that uses SNR frequency based distortion measures is also presented to assess the degradation effect of an imperfect noise estimator.

This talk is part of the COMMSP Seminar series.

This talk is included in these lists:

• 503
• Andrea Picciau's list
• COMMSP & CP list
• COMMSP Seminar
• Featured talks
• SAP Talks

Note that ex-directory lists are not shown.