http://mediafire.com

http://rapidshare.de

--- (Edited on 2/7/2009 12:43 am [GMT-0600] by nsh) ---

Hi Nsh,

Thanks again for all the info. I have uploaded the file to:

http://www.mediafire.com/?sharekey=622d284a325357dbd1014a7a667fa2b4e04e75f6e8ebb871

The test scripts I run were:

/source/PocketSphinx/pocketsphinx-0.5.1/scripts/pocketsphinx_myStuff (for using my own acoustic model at myTest1/).

/source/PocketSphinx/pocketsphinx-0.5.1/scripts/pocketsphinx_myStuff_adapt (for using the adapted acoustic model at myAdapt/).

The names to be recognized are in  /source/PocketSphinx/myTest1/etc/myTest1_train.transcription

BTW, please note that I had moved the dictionary file from /source/sphinxInfo/trainInfo/etc/7677.* to /source/PocketSphinx/ so the script may need to be updated.

I am not sure if I missed anything. If I do, please do not hesitate to let me know. Again, thank you so much for your help!

regards,

Jimmy

--- (Edited on 2/7/2009 1:44 pm [GMT-0600] by jcwang) ---

Hm, I suppose you just made a mistake, for example your transcription file misses utterance ids. Adaptation goes well for me. Or probably your sphinxtrain is old. Here is the result:

Before:

TOTAL Words: 18 Correct: 14 Errors: 7

TOTAL Percent correct = 77.78% Error = 38.89% Accuracy = 61.11%

TOTAL Insertions: 3 Deletions: 0 Substitutions: 4

After:

TOTAL Words: 18 Correct: 16 Errors: 4

TOTAL Percent correct = 88.89% Error = 22.22% Accuracy = 77.78%

TOTAL Insertions: 2 Deletions: 0 Substitutions: 2

Check my files here:

http://www.mediafire.com/?jerg1xddz4

--- (Edited on 2/7/2009 6:30 pm [GMT-0600] by nsh) ---

Hi Nsh,

Thanks for your response and help. I cannot access the file at: 

http://www.mediafire.com/?jerg1xddz4

--- (Edited on 2/8/2009 12:44 am [GMT-0600] by jcwang) ---

Ups, sorry

http://www.mediafire.com/download.php?jerg1xddz4s

--- (Edited on 2/8/2009 8:22 am [GMT-0600] by nsh) ---

Hi Nsh,

Thank you very much for your help and information! I really appreciate it!

regards,

Jimmy

--- (Edited on 2/9/2009 6:44 pm [GMT-0600] by jcwang) ---

Hi everybody

I try to run pocketsphinx on ubuntu with my data training .but when I training my data with SphinxStrain, my result files :feat.params,mdef,means,mixweights,noisedict,sendump,transition_matrices and variances .

But in pocketsphinx real data need mdef :bin file and not need file mixweights.My file training : mixweights:4.7mb mdef:267kb but in test data of pocketsphinx mdef:1.5mb

I try to rename my mixweights  to mdef .but error:

INFO: mdef.c(520): Reading model definition: /usr/local/share/pocketsphinx/model/hmm/en_US/hub4wsj_sc_8k/mdef
FATAL_ERROR: "mdef.c", line 539: Version error: Expecing 0.3, but read s3

I dont know how to make it run  so I really need your help .

Thankyou so much

[email protected]

--- (Edited on 5/18/2010 3:31 pm [GMT-0500] by Visitor) ---

> I try to run pocketsphinx on ubuntu with my data training .but when I training my data with SphinxStrain, my result files :feat.params,mdef,means,mixweights,noisedict,sendump,transition_matrices and variances.

You have everything needed to run pocketsphinx here. You can point model folder with -hmm option.

> But in pocketsphinx real data need mdef :bin file and not need file mixweights.

Pocketsphinx can work both with mdef or mdef.bin. You can convert between mdef and mdef.bin with command pocketsphinx_mdef_convert. There is no need to do conversion though. Mixture weights should be missing in model folder when you converted mixture weights to sendump file. If you have sendump, you don't need mixture weights. If you have mixture_weights you don't need sendump.

> I try to rename my mixweights  to mdef .but error:

You shouldn't do that. Just use model as it is.

--- (Edited on 5/19/2010 00:43 [GMT+0400] by nsh) ---

> You have everything needed to run pocketsphinx here. You can point model folder with -hmm option.

Can you tell me more details.I dont understand that.

When I run pocketSphinx on Win32 , I copy these model file to pocketsphinx/bin/debug and run pocketSphinx_continuous .

But in PocketSphinx on Ubuntu, I see some diffirent and I dont know what to do.

In PocketSphinx folder ,have 2 folder "hmm" and "lm" .I move my model files to hmm and my .dic file to lm replace to en_US/hub4wsj_sc_8k but I dont know what hub4.5000.DMP and wsj0vp.5000.DMP files  are. Can you tell me how to make it and if dont have it ,  the program can run correctly?

And the last one , could you  please give me the tutorial of how to built a pocketsphinx on Ubuntu with own data.

Thank you so much

Nguyen

--- (Edited on 5/19/2010 1:30 pm [GMT-0500] by Visitor) ---

> When I run pocketSphinx on Win32 , I copy these model file to pocketsphinx/bin/debug and run pocketSphinx_continuous .

It's error to copy model files. You need to understand how pocketsphinx configuration works. Pocketsphinx_continuous has command line options. They modify pocketsphinx_continuous behavior. If you provide option -dict <file.dict>, pocketsphinx uses this file.dict as a dictionary. If you add option -hmm <some_folder>, pocketsphinx will use model from this folder.

All you need to do is to provide pocketsphinx correct options:

pocketsphinx_continuous -dict <your.dict> -hmm <your_model_folder> -jsgf <your grammar>

Say you have grammar in the folder model_parameters/myam.cd_cont_1000. You need to run pocketsphinx with the following argument then:

pocketsphinx_continuous -hmm model_parameters/myam.cd_cont_1000

Folder path could be relative or absolute.

> but I dont know what hub4.5000.DMP and wsj0vp.5000.DMP files

Read

http://cmusphinx.sourceforge.net/wiki/languagemodelhowto

> And the last one , could you  please give me the tutorial of how to built a pocketsphinx on Ubuntu with own data.

http://cmusphinx.sourceforge.net/html/tutorial.html

--- (Edited on 5/19/2010 22:45 [GMT+0400] by nsh) ---