[hts-users:03210] Re: HSMMAlign for state level alignment and HMGenS for synthesis with aligned duration
- Subject: [hts-users:03210] Re: HSMMAlign for state level alignment and HMGenS for synthesis with aligned duration
- From: Keiichiro Oura <uratec@xxxxxxxxxxxxxxx>
- Date: Mon, 26 Mar 2012 02:55:06 +0900
- Cc: uratec <uratec@xxxxxxxxxxxx>
- Delivered-to: hts-users@xxxxxxxxxxxxxxx
- Dkim-signature: v=1; a=rsa-sha256; c=relaxed/relaxed; d=gmail.com; s=20120113; h=mime-version:sender:in-reply-to:references:date :x-google-sender-auth:message-id:subject:from:to:cc:content-type :content-transfer-encoding; bh=fSzfICUKQJYf7ydcnXBC4x0Y5suKlv6tRdxVJWZ7HJY=; b=lZPUdPAHcWeeUw17ocVZiHkqXUcB7eeK2X3QS0l2QCjt9u06XNzbeWGnhC0y5vBjJ5 BqTMGU1mHNABab9RUUwQyGVdZXYQH0J3qFjxp5IZX334qMRcN/3nM7zV02WD4XZfgPj9 XKB7nFYfAOa+avxknKfvNQ9N0iZCauIVSukI/JVNybDAMvClEYz+ZxN8OJcCS6ys7Pfo 47NInSYggbHdMizz9MJE6I++8F/UYgDHHHa2pM0L1KR3TGPV5oRd1PIrLFn8Xw+NBuJs tNEppkO+g6cmpp02rB9QukPlzvY8IWGuAti+ve/KGln9GzsUMrA9auhMwbuoJizFI/6x Wfpw==
Hi,
When you use '-s' option, state duration model is not required.
Regards,
Keiichiro Oura
2012/3/26 Heamin Lee <oasistony@xxxxxxxxxxx>
>
> In the generating speech parameter sequences (speaker independent), the error is below.
>
> I changed demo scripts about HMGenS from -e option to -s option and set $spkrPat to "*/*_%%%_*"
>
>
>
> HMGenS -A -B -C configs/syn.cnf -D -T 1 -S data/scp/gen.scp -t 1500 100 5000 -h */*_%%%_* -s -c 0 -H models/qst001/ver1/cmp/re_clustered_all.mmf -N models/qst001/ver1/dur/re_clustered_all.mmf -M gen/qst001/ver1/SI/0 models/qst001/ver1/cmp/tiedlist models/qst001/ver1/dur/tiedlist
>
>
>
> HTK Configuration Parameters[24]
>
> Module/Tool Parameter Value
>
> # CDGV TRUE
>
> # GVOFFMODEL StrVec 2 sil sp
>
> # OPTKIND NEWTON
>
> # GVWEIGHT 1
>
> # HMMWEIGHT 1
>
> # STEPDEC 0.500000
>
> # STEPINC 1.200000
>
> # STEPINIT 1
>
> # MINEUCNORM 0.010000
>
> # GVEPSILON 0.000100
>
> # MAXGVITER 50
>
> # GVHMMLIST ../gv/qst001/ver1/tiedlist
>
> # GVMODELMMF ../gv/qst001/ver1/clustered_all.mmf
>
> # USEGV TRUE
>
> # EMEPSILON 0.000100
>
> # MAXEMITER 20
>
> # WINDIR ../data/win
>
> # WINFN StrVec 3 mgc.win1 mgc.win2 mgc.win3 StrVec 3 lf0.win1 lf0.win2 lf0.win3 StrVec 3 bap.win1 bap.win2 bap.win3
>
> # PDFSTREXT StrVec 3 mgc lf0 bap
>
> # PDFSTRORDER IntVec 3 40 1 5
>
> # PDFSTRSIZE IntVec 3 1 3 1
>
> # USEALIGN TRUE
>
> # NATURALWRITEORDER TRUE
>
> # NATURALREADORDER TRUE
>
>
>
> HGen ML Generating: Transitions Durations Spaces Mixtures Parameters
>
>
>
> HMMSet is SHARED
>
> 110824 Logical/87000 Physical Models Loaded, VecSize=138
>
> 1 MMF input files
>
> PdfStream [1]:
>
> #streams: 1 (vSize=120)
>
> #order: 40
>
> file ext: mgc
>
> 1-th window: ../data/win/mgc.win1
>
> 2-th window: ../data/win/mgc.win2
>
> 3-th window: ../data/win/mgc.win3
>
> PdfStream [2]:
>
> #streams: 3 (vSize=3)
>
> #order: 1
>
> file ext: lf0
>
> 1-th window: ../data/win/lf0.win1
>
> 2-th window: ../data/win/lf0.win2
>
> 3-th window: ../data/win/lf0.win3
>
> PdfStream [3]:
>
> #streams: 1 (vSize=15)
>
> #order: 5
>
> file ext: bap
>
> 1-th window: ../data/win/bap.win1
>
> 2-th window: ../data/win/bap.win2
>
> 3-th window: ../data/win/bap.win3
>
>
>
> Generating Label tiedlist
>
> ERROR [+6510] LOpen: Unable to open label file ../models/qst001/ver1/dur/tiedlist.lab
>
> FATAL ERROR - Terminating program HMGenS
>
> Error in HMGenS -A -B -C configs/syn.cnf -D -T 1 -S data/scp/gen.scp -t 1500 100 5000 -h "*/*_%%%_*" -s -c 0 -H models/qst001/ver1/cmp/re_clustered_all.mmf -N models/qst001/ver1/dur/re_clustered_all.mmf -M gen/qst001/ver1/SI/0 models/qst001/ver1/cmp/tiedlist models/qst001/ver1/dur/tiedlist
>
>
>
>
>
> From: ura228@xxxxxxxxx [mailto:ura228@xxxxxxxxx] On Behalf Of Keiichiro Oura
> Sent: Saturday, March 24, 2012 4:48 PM
> To: hts-users@xxxxxxxxxxxxxxx
> Cc: uratec
> Subject: [hts-users:03208] Re: HSMMAlign for state level alignment and HMGenS for synthesis with aligned duration
>
>
>
> Hi,
>
>
>
> What was the problem that caused in HMGenS step?
>
>
>
> Regards,
>
> Keiichiro Oura
>
>
>
> 2012/3/23 Heamin Lee <oasistony@xxxxxxxxxxx>
>
> Hello,
>
> I want evaluate the performance by Objective evaluation in HTS papers.
>
> So I've tried HMGenS with '-s' option to use state alignment for duration.
>
> And I used HSMMAlign with-f option to generate state level alignment..
>
> The output is below,
>
> 0 1650000 x^x-sil+S=AX@x_x/A:0/B:x-x-&x-x|x/C:0/D:0/E:x+&x/F:0/H:12=33[2] x^x-sil+S=AX@x_x/A:0/B:x-x-&x-x|x/C:0/D:0/E:x+&x/F:0/H:12=33
>
> 1650000 1700000 x^x-sil+S=AX@x_x/A:0/B:x-x-&x-x|x/C:0/D:0/E:x+&x/F:0/H:12=33[3]
>
> 1700000 1750000 x^x-sil+S=AX@x_x/A:0/B:x-x-&x-x|x/C:0/D:0/E:x+&x/F:0/H:12=33[4]
>
> 1750000 1800000 x^x-sil+S=AX@x_x/A:0/B:x-x-&x-x|x/C:0/D:0/E:x+&x/F:0/H:12=33[5]
>
> 1800000 1950000 x^x-sil+S=AX@x_x/A:0/B:x-x-&x-x|x/C:0/D:0/E:x+&x/F:0/H:12=33[6]
>
> 1950000 2000000 x^sil-S+AX=UW@1_2/A:0/B:2-1-&1-33|AX/C:2/D:0/E:2+&12/F:2/H:12=33[2] x^sil-S+AX=UW@1_2/A:0/B:2-1-&1-33|AX/C:2/D:0/E:2+&12/F:2/H:12=33
>
> 2000000 2100000 x^sil-S+AX=UW@1_2/A:0/B:2-1-&1-33|AX/C:2/D:0/E:2+&12/F:2/H:12=33[3]
>
> 2100000 2450000 x^sil-S+AX=UW@1_2/A:0/B:2-1-&1-33|AX/C:2/D:0/E:2+&12/F:2/H:12=33[4]
>
> 2450000 2650000 x^sil-S+AX=UW@1_2/A:0/B:2-1-&1-33|AX/C:2/D:0/E:2+&12/F:2/H:12=33[5]
>
> 2650000 3000000 x^sil-S+AX=UW@1_2/A:0/B:2-1-&1-33|AX/C:2/D:0/E:2+&12/F:2/H:12=33[6]
>
> 3000000 3550000 sil^S-AX+UW=L@2_1/A:0/B:2-1-&1-33|AX/C:2/D:0/E:2+&12/F:2/H:12=33[2] sil^S-AX+UW=L@2_1/A:0/B:2-1-&1-33|AX/C:2/D:0/E:2+&12/F:2/H:12=33
>
>
>
> However, when I used this file for a input of HMGenS with '-s' option, a problem came up.
>
> And I also tried HMGenS with '-m' option with a file that is not state level alignment, I could get right result.
>
> In this situation, I want to know that the output from HSMMAlign with -s option is appropriate format for a input of HMGenS ?
>
>
>
> 오류! 파일 이름이 지정되어 있지 않습니다.
>
>
- Follow-Ups
-
- [hts-users:03212] Re: HSMMAlign for state level alignment and HMGenS for synthesis with aligned duration, Heamin Lee
- References
-
- [hts-users:03205] HSMMAlign for state level alignment and HMGenS for synthesis with aligned duration, Heamin Lee
- [hts-users:03208] Re: HSMMAlign for state level alignment and HMGenS for synthesis with aligned duration, Keiichiro Oura
- [hts-users:03209] Re: HSMMAlign for state level alignment and HMGenS for synthesis with aligned duration, Heamin Lee