Efficient Noise Suppression for Robust Speech Recognition
Cited 0 times inCited 0 times in
- Efficient Noise Suppression for Robust Speech Recognition
- Lee, Kangyeoul
- Jang, Gil-Jin
- Issue Date
- Graduate School of UNIST
- This thesis addresses the issues of single microphone based noise estimation technique for speech recognition in noise environments. A lot of researches have been performed on the environmental noise estimation, however most of them require voice activity detector (VAD) for accurate estimation of noise characteristics. I propose two approaches for efficient noise estimation without VAD. The first approach aims at improving the conventional quantile-based noise estimation (QBNE). I fostered the QBNE by adjusting the quantile level (QL) according to the relative amount of added noise to the target speech. Basically, we assign two different QLs, i.e., binary levels, according to the measured statistical moment of log scale power spectrum at each frequency. The second approach is applying dual mixture parametric model in computing likelihoods of speech and non-speech classes. I used dual Gaussian mixture model (GMM) and Rayleigh mixture model (RMM) for the likelihoods. From the assumption that speech is generally uncorrelated to the environmental noises, the noise power spectrum can be estimated by using each mixture model parameter of speech absence class.
I compared the proposed methods with the conventional QBNE and minimum statistics based method on a simple speech recognition task in various signal-to-noise ratio (SNR) levels. Based on the experimental results, the proposed methods are shown to be superior to the conventional methods.
- Electrical Engineering
- ; Go to Link
- Appears in Collections:
- Files in This Item:
Efficient Noise Suppression.pdf
can give you direct access to the published full text of this article. (UNISTARs only)
Show full item record
Items in DSpace are protected by copyright, with all rights reserved, unless otherwise indicated.