Timeliner
From SpeechWiki
ruby codes
/workspace/ifp-32-2/hasegawa/data/multimodal/nonspeech/FODAVA/timeliner (the feature file opened is currently hardcoded , update that line before using)
set up the environment
a) Ubuntu (verified with Xiaodan's 9.04, Camille's 64bit 8.10, Mark's ?8.04.)
b) update all security updates, aptitude update
c) install packages with aptitude(or apt-get) and gem (some back and forth to discover dependencies)
aptitude install ruby ruby1.8-dev rubygems1.8
aptitude install gem; gem update
aptitude install mesa-common-dev libglu1-mesa-dev freeglut3-dev imagemagick libmagickcore-dev libmagickwand-dev (may need change order)
gem install rake mkrf rmagick RubyInline ruby-opengl rspec ZenTest
Memos:
locate: find a file on local disk
apt-file find ruby.h: find packages that generate ruby.h
feature generating (to be detailed)
/workspace/ifp-32-2/hasegawa/xzhuang2/AED2009/tmp/forCamille
A. To make a feature file: HCopy -C $ConfigFile AIT_20061020_AmarkIII_1.ch4.wav $ResultFile
B. To view a feature file: HList -h $ResultFile
| ConfigFile | ResultFile | Description |
|---|---|---|
| HCopy_FB.cfg | AIT_20061020_AmarkIII_1.ch4.feaFB | narrow-band filterbank (window 25ms, step 10ms), 78-dim |
| HCopy_MFCC.cfg | AIT_20061020_AmarkIII_1.ch4.feaMFCC | MFCC (window 25ms, step 10ms), 78-dim |
| HCopy_FB_w6ms_o2ms.cfg | AIT_20061020_AmarkIII_1.ch4.feaFB_w6ms_o2ms | wide-band filterbank (window 6ms, step 2ms), 78-dim |
Note:
AIT_20061020_AmarkIII_1.ch4.wav wavform
AIT_20061020_AmarkIII_1.ch4.annhtk 92 dims (14 dim decorrelated event-classifier neural network output + 78 dim filter bank parameters)
1) For each 78 dim filter bank parameters (or MFCCs), the first 26 dims are original, the second and third 26 dims are first-order and second-order regression coefficients derived from the first 26 dims respectively.
