[Subject Prev][Subject Next][Thread Prev][Thread Next][Date Index][Thread Index]

[hts-users:00701] Re: hts_engine: Phoneme Symbols

Hi Heiga ZEN,
 
Here is a part of my label file.
 
XXX_XXX-sil+h=@/A:x
XXX_sil-h+@=ng^/A:2
sil_h-@+ng^=s/A:2
h_@-ng^+s=a/A:2
@_ng^-s+a=m/A:1
ng^_s-a+m=u/A:1
s_a-m+u=t^/A:1
...
and so on.
 
and here is a my trace file.
 
sampring frequency                     -> 8000(Hz)
frame period                           -> 40(point) 5.00(msec)
use state alignment for duration       -> 0
use phoneme alignment for duration     -> 0
all-pass constant                      -> 0.420000
postfiltering coefficient              -> 0.400000
control duration parameter             -> 0.000000
multilply f0                           -> 1.000000
add f0                                 -> 0.000000
voiced/unvoiced threshold              -> 0.500000
specified utterance length             -> 0.000000(sec.)
number of HMMs        -> 4
number of HMM states  -> 20
length of this speech -> 0.430 sec. (86 frames)
1: XXX_XXX-sil+h=@/A:x
            duration -> 1105
  2-state : spectrum -> 1     f0 -> 1      0.000--0.015(sec)   3(frame)   unvoiced
  3-state : spectrum -> 20    f0 -> 586    0.015--0.055(sec)   8(frame)   unvoiced
  4-state : spectrum -> 6     f0 -> 22     0.055--0.135(sec)  16(frame)   unvoiced
  5-state : spectrum -> 741   f0 -> 772    0.135--0.205(sec)  14(frame)   unvoiced
  6-state : spectrum -> 37    f0 -> 389    0.205--0.220(sec)   3(frame)   unvoiced
2: XXX_sil-h+@=ng^/A:2
            duration -> 578
  2-state : spectrum -> 15    f0 -> 126    0.220--0.230(sec)   2(frame)   unvoiced
  3-state : spectrum -> 14    f0 -> 40     0.230--0.245(sec)   3(frame)   unvoiced
  4-state : spectrum -> 854   f0 -> 495    0.245--0.280(sec)   7(frame)   unvoiced
  5-state : spectrum -> 37    f0 -> 6      0.280--0.290(sec)   2(frame)   unvoiced
  6-state : spectrum -> 568   f0 -> 234    0.290--0.295(sec)   1(frame)   voiced
3: sil_h-@+ng^=s/A:2
            duration -> 374
  2-state : spectrum -> 209   f0 -> 1002    0.295--0.305(sec)   2(frame)   voiced
  3-state : spectrum -> 579   f0 -> 145    0.305--0.320(sec)   3(frame)   voiced
  4-state : spectrum -> 23    f0 -> 298    0.320--0.330(sec)   2(frame)   voiced
  5-state : spectrum -> 562   f0 -> 683    0.330--0.345(sec)   3(frame)   voiced
  6-state : spectrum -> 45    f0 -> 183    0.345--0.355(sec)   2(frame)   voiced
4: h_@-ng^+s=a/A:2
            duration -> 804
  2-state : spectrum -> 107   f0 -> 370    0.355--0.385(sec)   6(frame)   voiced
  3-state : spectrum -> 638   f0 -> 502    0.385--0.400(sec)   3(frame)   voiced
  4-state : spectrum -> 81    f0 -> 287    0.400--0.415(sec)   3(frame)   voiced
  5-state : spectrum -> 626   f0 -> 493    0.415--0.420(sec)   1(frame)   voiced
  6-state : spectrum -> 410   f0 -> 489    0.420--0.430(sec)   2(frame)   voiced
 

I have 248 number of HMMs in the label file. The next line should be '@_ng^-s+a=m/A:1'.

This happens to me both on Linux and Windows (on every label file like this one).

I use '@' and '@@' for phoneme, not for class separator. Using the symbols as class separator work fine for me.

I am a HTS newbie, sorry i cannot give you a debugger information.

 

 

Regards,

Nipon Chi.

 





> Date: Thu, 31 May 2007 11:44:32 +0900
> From: zen@xxxxxxxxxxxxxxx
> Subject: [hts-users:00695] Re: hts_engine: Phoneme Symbols
> To: hts-users@xxxxxxxxxxxxxxx
>
> Hi,
>
> Nipon Chinathimatmongkhon wrote (2007/05/31 9:43):
>
> > I have used '@' and '@@' to represent phoneme.
> > Every time I use hts_engine, I always stop at '@@' and '@' phone.(check from trace file)
> > Is these symbols usable?
>
> Hmm, it's weird because we are using '@' in our English label format.
> For example
>
> pau^pau-ao+th=er@1_2/A:0_0_0/B:1-1-2@1-2&1-7#1-4$1-3!0-2;0-4|ao/C:0+0+1/D:0_0/E:content+2@1+5&1+2#0+3/F:in_1/G:0_0/H:7=5@1=2|L-L/I:7=3/J:14+8-2
>
> So it should work.
> What kind of error do you receive? Segmentation fault?
> Please run debugger and give us more information about this problem
>
> > And is there any other symbol I have to avoid?
>
> You should avoid '*' and '?' because they are used as special characters for pattern matching.
> In HTK, '%' is also used as a special character, so it should also be avoided.
>
> 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
> ------------------------------------------------
>


Change is good. See what’s different about Windows Live Hotmail. Check it out!