Arabic acoustic model

[khaledel]

Hi,

I'm trying to create an arabic acoustic model using SphinxTrain. I've create the files like (dictionnary, transcription ...).

When i try to run the SphinxTrain to create the acoustic model using the command : runAll.pl, i got this error :

(unsigned)(c+1) <= 256 error

File : isctype.c
Line : 56
Line 68

Do someone have any idea, why i'm getting this error ???

I've already created a french acoustic model using the same soft and i didn't any errors.

Thx for your help.

[Marina Abdel-Malak]

Hello Nickolay,

Thanks for replying, i've resolved the problem.

well i'm using sphinx 4 and i've made a program that converts each arabic letter to an english one. and i took care of the spacing issue.. anyway thanks :)

Marina

[Nickolay V. Shmyrev]

This is a bug in Sphinxtrain. I've just fixed few tolower calls in trunk with the following patch:

Index: src/libs/libcommon/lts.c

--- src/libs/libcommon/lts.c (revision 9059)
+++ src/libs/libcommon/lts.c (working copy)
@@ -81,7 +81,7 @@
word = ckd_salloc((char *)in_word);
if (!r->letter_table)
for (i = 0; i < strlen(word); ++i)
- word[i] = tolower(word[i]);
+ word[i] = tolower((int)word[i]);

 /* Fill in out_phones structure as best we can. */
 maxphones = strlen(word) + 10;

Index: src/libs/libio/corpus.c

--- src/libs/libio/corpus.c (revision 9059)
+++ src/libs/libio/corpus.c (working copy)
@@ -233,10 +233,10 @@
strcpy(b_lc, b);

 for (i = 0; i < strlen(a_lc); i++)

• a_lc[i] = tolower(a_lc[i]);

• a_lc[i] = tolower((int)a_lc[i]);

  for (i = 0; i < strlen(b_lc); i++)
  - b_lc[i] = tolower(b_lc[i]);
  + b_lc[i] = tolower((int)b_lc[i]);

  return strcmp(a_lc, b_lc);
  }

But probably some more left. You need to check warnings during compilation after update.

[khaledel]

hey,

what exactly should i do to repare this bug ???

thx in advance

[khaledel]

For more information, the pbm i'm having is in : MODULE: 20 Training Context Independent models.

Phase 3: Forward-Backward

Baum welch starting for 1 Gaussian(s), iteration: 1 (1 of 1)

bw Log File
completed

Only 0 parts of 1 of Baum Welch were successfully completed

Parts 1 failed to run!

Training failed in iteration 1

---

I really need your help !!!!!!!!!!!!!!!

[eliasmajic]

Hi Khaledel

I recall having a error at that part a while back that was due to me not having a matching word in the dictionary for words in my transcription files. Take a look at the html file, it has much more detailed info on what is going wrong and would help if you posted it so we can see it. Post the html and the log file that it links to where the error is occuring.

[Marina Abdel-Malak]

hi,
i'm also new to sphinx .i'm doing my bachelor project and my topic is arabic speech recognition adaptation and i have a problem in which sphinx train doesn't understand the Arabic letters in the dictionary file. and in the command prompt it displays that the sentence exists in the transcription but not in the dictionary and about the phonemes it exists in the phone list and not in the transcription...
here's part of my dictionary :

سامحنى S A M I H- N I
ارجوك A- R G O O K
بزيادة B I Z Y A A D A
كفاية K I F A A Y A
ورينى W A R I N I
بطاقة B I T A K A
الشخصية E L SS A X S' I Y A

and a part of the transcription

<s>ارجوك</s> ( aSentence120)
<s>سامحنى</s> ( aSentence121)
<s>بزيادة</s> ( aSentence122)
<s>وريني بطاقة الشخصية</s> ( aSentence018)

Hope u can help me :)
thanks in advance

[Nickolay V. Shmyrev]

Hello Marina

I tried to reproduce your issue but everything works fine with the latest pocketsphinx. There are only whitespace issues you need to check:

<s>ارجوك</s> ( aSentence120)

put spaces after <s>, put space before </s>, don't put space after open brace and make sure you don't have any spaces after closing brace:

<s> ارجوك </s> (aSentence120)

After that everythign should just work.

[net_king]

me too i have the same problem with sphinx train !!!