Force Align

[Joel Madhan Raj]

Hi,

I have tried force align and got two files named in folder falignout,

  1) test.alignedfiles
  2) test.alignedtranscripts

This is the sample line i have got from aligned transcript.
<s> ENJOYING HIS <sil> REGULAR(2) ACTIVITIES(2) </s>

Then i generated new training material based on above two files and when i started to decode it shows error in decoding part such as

FATAL_ERROR: "e:\tutorial\sphinx3\src\libs3decoder\libdict\dict.c", line 204: Missing base word for: ACTIVITIES(2)

I have the word ACTIVITIES(2) in dictionary.

Is there is any way to solve the problem while decoding.

Thanks in advance

with regards,
E.Joel Madhan Raj

[Nickolay V. Shmyrev]

It tells your dictionary misses ACTIVITIES, not ACTIVITIES(2). Please check the dictionary again. Also check that your dictionary has not inconsistent newlines (LF vs CR+LF, they could be the problem.

[Joel Madhan Raj]

Hi,

Thanks for ur reply.I created the dictionary with ACTIVITIES,ACTIVITIES(2) and tried to decode but it shows an error

ERROR: "e:\tutorial\sphinx3\src\libs3decoder\libdict\wid.c", line 148: LM word 'ACTIVITIES(2)' is an alternative pronunciation in dictionary

and gets corrupted. I tired the other way by deleting all the multiple pronunciation and started to decode and i got the output. But the accuracy is even worse then the one without force align.

Please tell the process of force aligning in training and decoding. Is there any documentation regarding the force align running in sphinx3.
Thanks in advance.

with regards,
E.Joel Madhan Raj