Re: Convolutive feature extraction (Searching for papers) (Paris Smaragdis )

Subject: Re: Convolutive feature extraction (Searching for papers)
From:    Paris Smaragdis  <paris@xxxxxxxx>
Date:    Fri, 13 Oct 2006 10:38:29 -0400

I wrote these papers that probably describe what you want:

http://www.merl.com/reports/docs/TR2004-094.pdf
http://www.merl.com/reports/docs/TR2007-002.pdf

Also look up Tuomas Virtanen's papers on convolutive NMF.  There are  
also some 2D extensions by Schmidt & Morup and by Fitzgerald which  
are best suited for finding features on constant-q spectra.

Paris



On Oct 12, 2006, at 2:53 PM, ÃÖÃæȯ wrote:

> Dear List,
>
> Recently, I started to look into extracting dynamic speech features  
> from multiple frames(convolutive feature).
>
> For example, as a input vector of HMM, the delta and delta-delta of  
> mfcc is computed from cepstrum which is extracted from single frame.
> Contrary to this approach, convolutive feature is trying to extract  
> features from multiple frames at one time.
>
> Firstly,I was searching recent papers using 'speech, dynamic,  
> convolutive, multi frame, feature, etc' as a keyword.
> But I couldn't find papers related to that field. It was hard to  
> find out how much work was done on this field.
> Would you please help me to find recent work(maybe a paper) with  
> multi-frame features in speech?
> I'd love to hear some advices from you.
>
> Thanks,
> Choong
>
>
> ====================================================
> Choong Hwan Choi
> Ph.D. candidate
> Computational NeuroSystems Laboratory
> Department of BioSystems
> Korea Advanced Institude of Science and Technology
> Tel: +82-42-869-8492
> E-mail: imchwan@xxxxxxxx
> ====================================================
>