|
dear all, plz give example 1.what kind of features additionaly can i use to prevent the convergence ? 2. when can i used the HMM trained with dynamic features and in the HMGens i draw the spectra without it ... in which case ?. i need to study the effect of dynamic features on the generated spectra ! Thanks > Date: Mon, 8 Oct 2007 12:54:13 +0000 > From: sacha@xxxxxxxxxxxxxxxxx > Subject: [hts-users:00828] Re: Static And Dynamic Features > To: hts-users@xxxxxxxxxxxxxxx > > > Hi Tamer; > > If you train HMMs with different features, chances are that the EM > algorithm will not converge > to the same solution at the end of the training. So both methods are > *not* equivalent, > and the applicable method depends on the purpose of your experiment. > > Regards; > -*- Sacha -*- > > Tamer Fares wrote: > > > > > > ------------------------------------------------------------------------ > > > > > > dear All > > > > > > i asked question to study the presence and absence of dynamic > > features on the spectra > > > > can we consider the method mentioned in this mail as an equivalent > > method to the one if i tried to train the HMM without the dynamic > > > > i mean that if i tried to draw the spectral using the way that heiga > > mentioned in this mail ; it will be like the spectral if i was used > > the cmp without the dynamic features before the training > > > > is this be practical if i study the effect of (presence and abscence > > ) of dynamic features with HMM trained with dynamic feature > > > > or should i train the HMM without the dynamic features to draw the > > spectral and evaluate the difference > > > > Tamer Fares > > > > > > > Date: Tue, 2 Oct 2007 14:58:50 +0900 > > > From: zen@xxxxxxxxxxxxxxx > > > Subject: [hts-users:00815] Re: Static And Dynamic Features > > > To: hts-users@xxxxxxxxxxxxxxx > > > > > > Hi, > > > > > > Tamer Fares wrote (2007/09/24 17:47): > > > > > > > so plz let me know if i will use the HMM trained with dynamic > > features > > > > to synthesis speech without the dynamic features > > > > > > > > i mean how to synthesis voice without the usage od dynamic features > > > > > > > > what steps to follow to make this ? > > > > > > Modify configs/syn.cnf from > > > > > > NATURALREADORDER = T > > > NATURALWRITEORDER = T > > > USEALIGN = T > > > PDFSTRSIZE = "IntVec 2 1 3" > > > PDFSTRORDER = "IntVec 2 25 1" > > > PDFSTREXT = "StrVec 2 mcp lf0" > > > WINFN = "StrVec 3 mcp.win1 mcp.win2 mcp.win3 StrVec 3 lf0.win1 > > lf0.win2 lf0.win3 " > > > WINDIR = > > /common_work/zen/HTS-2.0-release/HTS-demo_Nitech-ATR503-M001/data/win > > > > > > to > > > > > > NATURALREADORDER = T > > > NATURALWRITEORDER = T > > > USEALIGN = T > > > PDFSTRSIZE = "IntVec 2 1 3" > > > PDFSTRORDER = "IntVec 2 75 1" > > > PDFSTREXT = "StrVec 2 mgc lf0" > > > WINFN = "StrVec 1 mgc.win1 StrVec 3 lf0.win1 lf0.win2 lf0.win3 " > > > WINDIR = > > /common_work/zen/HTS-2.0-release/HTS-demo_Nitech-ATR503-M001/data/win > > > > > > The above modification means that constraints between static and > > dynamic features are not used in the speech parameter generation > > algorithm. > > > After running HMGenS, you obtain 75-dimensional mcp sequences. > > > First 25 dimensions of this 75-dimensional mcp sequences are > > static features, so pick up them using bcp of SPTK as follows: > > > > > > bcp +f -l 75 -L 25 -s 0 -e 24 -S 0 alice01.mcp > > > alice01_wo_dynamic.mcp > > > > > > Please refer to the SPTK's reference manual for details. > > > Use alice01_wo_dynamic.mcp instead of alice01.mcp in the waveform > > synthesis stage. > > > > > > Regards, > > > > > > Heiga ZEN (Byung Ha CHUN) > > > > > > -- > > > ------------------------------------------------ > > > Heiga ZEN (in Japanese pronunciation) > > > Byung Ha CHUN (in Korean pronunciation) > > > > > > Department of Computer Science and Engineering > > > Nagoya Institute of Technology > > > Gokiso-cho, Showa-ku, Nagoya 466-8555 Japan > > > > > > http://www.sp.nitech.ac.jp/~zen > > > ------------------------------------------------ > > > > > > > > > ------------------------------------------------------------------------ > > Discover the new Windows Vista Learn more! > > <http://search.msn.com/results.aspx?q=windows+vista&mkt=en-US&form=QBRE> > > > > > > > > ------------------------------------------------------------------------ > > Explore the seven wonders of the world Learn more! > > <http://search.msn.com/results.aspx?q=7+wonders+world&mkt=en-US&form=QBRE> > > ______________________________________________________________________ > > This email has been scanned by the MessageLabs Email Security System. > > For more information please visit http://www.messagelabs.com/email > > ______________________________________________________________________ > > -- > Dr.Sacha Krstulovic - Research Engineer > Toshiba Research Europe Limited > Cambridge Research Laboratory > Speech Technology Group > 208 Science Park, Milton Road > Cambridge CB4 0GZ - United Kingdom > > > > ______________________________________________________________________ > This email has been scanned by the MessageLabs Email Security System. > For more information please visit http://www.messagelabs.com/email > ______________________________________________________________________ Invite your mail contacts to join your friends list with Windows Live Spaces. It's easy! Try it! |