Menu

adapting acoustic model

Help
med
2012-03-02
2012-09-22
1 2 > >> (Page 1 of 2)
  • med

    med - 2012-03-02

    Hi,
    I've in "hub4wsjs_sc_8k" : amdigits.dic amdigits.fileids amdigits.txt
    amdigits.transcription and the .wav files

    when I run this command this is what I get:
    hyt-med@Hyt-Med:~/Desktop/hub4wsj_sc_8k$ sudo sphinx_fe -argfile
    hub4wsj_sc_8k/feat.params -samprate 16000 -c arctic20.fileids -di . -do . -ei
    wav -eo mfc -mswav yes
    password for hyt-med:
    INFO: cmd_ln.c(691): Parsing command line:
    sphinx_fe \
    -argfile hub4wsj_sc_8k/feat.params \
    -samprate 16000 \
    -c arctic20.fileids \
    -di . \
    -do . \
    -ei wav \
    -eo mfc \
    -mswav yes

    Current configuration:

    -alpha 0.97 9.700000e-01
    -argfile hub4wsj_sc_8k/feat.params
    -blocksize 2048 2048
    -build_outdirs yes yes
    -c arctic20.fileids
    -cep2spec no no
    -di .
    -dither no no
    -do .
    -doublebw no no
    -ei wav
    -eo mfc
    -example no no
    -frate 100 100
    -help no no
    -i
    -input_endian little little
    -lifter 0 0
    -logspec no no
    -lowerf 133.33334 1.333333e+02
    -mach_endian little little
    -mswav no yes
    -ncep 13 13
    -nchans 1 1
    -nfft 512 512
    -nfilt 40 40
    -nist no no
    -npart 0 0
    -nskip 0 0
    -o
    -ofmt sphinx sphinx
    -part 0 0
    -raw no no
    -remove_dc no no
    -round_filters yes yes
    -runlen -1 -1
    -samprate 16000 1.600000e+04
    -seed -1 -1
    -smoothspec no no
    -spec2cep no no
    -sph2pipe no no
    -transform legacy legacy
    -unit_area yes yes
    -upperf 6855.4976 6.855498e+03
    -verbose no no
    -warp_params
    -warp_type inverse_linear inverse_linear
    -whichchan 0 0
    -wlen 0.025625 2.562500e-02

    ERROR: "cmd_ln.c", line 787: Cannot open configuration file
    hub4wsj_sc_8k/feat.params for reading
    ERROR: "sphinx_fe.c", line 1099: Command line parsing failed

    I don't know what I'm missing,
    thank you for your help

     
  • hiyassat

    hiyassat - 2012-03-02

    point to the feat.params file

     
  • Nickolay V. Shmyrev

    I don't know what I'm missing,

    You miss the understanding of the following core concepts of the Unix system:

    1. Current working folder
    2. Relative file path specification
    3. File operations (open, read write)

    Because of that you are running the command from a wrong current folder you
    get the error message saying that file is not found.

     
  • med

    med - 2012-03-02

    yes I'm : hyt-med@Hyt-Med:~/Desktop/hub4wsj_sc_8k$
    feat.params exist in "hub4wsj_sc_8k" dir
    any other ideas?

     
  • med

    med - 2012-03-02

    I'm sorry, I find it :s
    is this means that the command succes? INFO: sphinx_fe.c(1043): Processing all
    remaining utterances at position 0

    thank you so much, and sorry again

     
  • Nickolay V. Shmyrev

    is this means that the command succes? INFO: sphinx_fe.c(1043): Processing
    all remaining utterances at position 0

    Yes

     
  • med

    med - 2012-03-03

    thank you.
    I'm using pocketsphinx but in the tutorial I read :
    ./mllr_solve \
    -meanfn /usr/local/share/sphinx3/model/hmm/hub4_cd_continuous_8gau_1s_c_d_dd/means \
    -varfn /usr/local/share/sphinx3/model/hmm/hub4_cd_continuous_8gau_1s_c_d_dd/variances \
    -outmllrfn mllr_matrix -accumdir
    I tried to change sphinx3 with pocketsphinx but in hmm there are three dir
    "en" "en_US" and "zh", and everyone contains "means" and "variances"...
    can you please tell me what to do now

     
  • Nickolay V. Shmyrev

    I'm not sure what tutorial are you talking about. Proper tutorial at http://c
    musphinx.sourceforge.net/wiki/tutorialadapt
    doesn't have anything similar. It mentions exact step by
    step howto on pocketsphinx model adatation.

     
  • med

    med - 2012-03-03

    I mean this paragraph:
    Creating transformation with MLLR

    MLLR transforms are supported by pocketsphinx and sphinx4. MLLR is a cheap
    adaptation method that is suitable when amount of data is limited. It's a good
    idea to use MLLR for online adaptation. If you want best accuracy you can
    combine MLLR adaptation with MAP adaptation below

    Next we will generate an MLLR transformation which we will pass to the decoder
    to adapt the acoustic model at run-time. This is done with the mllr_solve
    program:

    ./mllr_solve \
    -meanfn /usr/local/share/sphinx3/model/hmm/hub4_cd_continuous_8gau_1s_c_d_dd/means \
    -varfn /usr/local/share/sphinx3/model/hmm/hub4_cd_continuous_8gau_1s_c_d_dd/variances \
    -outmllrfn mllr_matrix -accumdir .

    Now, if you wish to decode with the adapted model, simply add -mllr
    mllr_matrix (or whatever the path to the mllr_matrix file you created is) to
    your pocketsphinx command line.

    I've nothing with the name sphinx3 in the path /usr/local/share/

    and my other question is that I've a lot of . wav files for the same word, I
    mean for example the word "Ilem" has ilem00.wav ....ilem09.wav, the same for
    the other words,
    tell me if this is how . fileids has to look like :
    ilem00
    ilem01
    .
    .
    .
    yen00
    yen01
    .
    .
    .
    and the .transcription:
    ILEM (Ilem00) (ilem01) .........(ilem09)
    YEN (Yen00) (Yen01)...........(Yen09)
    ...

    Sorry for all those questions, thank you

     
  • Nickolay V. Shmyrev

    mean this paragraph: Creating transformation with MLLR

    I've updated the tutorial. Check it again

    and the .transcription: ILEM (Ilem00) (ilem01) .........(ilem09)
    YEN (Yen00) (Yen01)...........(Yen09)

    This transcription file format is not correct. Each file should have one line
    of transcription. You need to have duplicated transcription lines for files
    with the same content.

     
  • med

    med - 2012-03-03

    Ah thank you it's working now :)

    you mean like this:
    ILEM (Ilem00)
    ILEM (Ilem01)
    .
    .
    .
    ILEM (Ilem09)
    and the same for the other words?

     
  • med

    med - 2012-03-03

    Sorry, i don't know what this mean:

    pocketsphinx_batch \
    -adcin yes \
    -cepdir wav \
    -cepext .wav \
    -ctl adaptation-test.fileids \
    -lm <your.lm> \
    -dict <your.dict> \
    -hmm <your_new_adapted_model> \
    -hyp adapation-test.hyp </your_new_adapted_model></your.dict></your.lm>

    word_align.pl adaptation-test.transcription adapation-test.hyp

    <your.lm> is the .lm that I had, but there is no ".dict", just ".dic", and do
    you mean by <your_new_adapted_model> this forlder "hub4wsj_sc_8kadapt" ?!! </your_new_adapted_model></your.lm>

    I think this part of the tutorial is really hard to understand, can you please
    help me, my goal is to creat an application bases on pocketsphinx and
    recognize the alphabet of my langage, but I'm not sure if I'm going on the
    right way....

     
  • ashwani

    ashwani - 2012-03-04

    pocketsphinx_batch \
    -adcin yes \
    -cepdir wav \
    -cepext .wav \
    -ctl adaptation-test.fileids \
    -lm <your.lm> \
    -dict <your.dict> \
    -hmm <your_new_adapted_model> \
    -hyp adapation-test.hyp</your_new_adapted_model></your.dict></your.lm>

    hiii medleg

    this part of tutorial tell you about testing of your adapted model.
    in your model directory make test folder and all other test files as er the
    tutorial and then copy paste the command as it is it will give your accuracy
    result of the adapted acoustic model.
    but change "your.lm" to language model path you have created from your
    arctic.text using tutorial "Building language model cmusphinx"...google it to
    get the link.
    same with "your.dic"
    and "your adapted model" ...if you run in your working directory i.e where
    your model resides....give the folder name there eg: hub4wsj_sc_8kadapt

    you will find the python file word_align in pocketsphinx folder you have
    downloaded.copy this also in your working directory.

     
  • med

    med - 2012-03-04

    Thank you so much ashwanijanghu, I really apprecite it :), it works great
    so now I've to read the next tutorial "Training Acoustic Model For CMUSphinx",
    is it the right step?
    thank you again

     
  • ashwani

    ashwani - 2012-03-04

    Thank you so much ashwanijanghu, I really apprecite it :), it works great
    so now I've to read the next tutorial "Training Acoustic Model For CMUSphinx",
    is it the right step?
    thank you again

    it depends medleg..
    if you want to adapt an existing model than go for adaptation but if you want
    to start from scratch then for that tutorial but it will be more troublesome
    ....better you adapt model and improve accuracy by providing better quality
    and exact configuration wav files....

     
  • med

    med - 2012-03-04

    I'm sorry, I'm lost, can you please tell me the steps to creat an application
    based on pocketsphinx to recognize numbers from 0-->9 in arabic, please help
    me because I read a lot about this and I did a lot of work but I think I'm not
    going in the right direction, and I have just 2 months to finish this work :s

    thank you

     
  • Nickolay V. Shmyrev

    I'm sorry, I'm lost, can you please tell me the steps to creat an
    application based on pocketsphinx to recognize numbers from 0-->9 in arabic,
    please help me because I read a lot about this and I did a lot of work but I
    think I'm not going in the right direction, and I have just 2 months to finish
    this work :s

    Sorry, I do not know what you has done already, what is your experience and
    what problem do you have right now. So I can't tell you much on what to do
    next.

     
  • med

    med - 2012-03-06

    okay, I'll explain what I did for now, I ve a data base of numbers & alphabet
    (about 30 person, everyone repear ten times every singles number and alphabet,
    I class them into a dir (.wav)), and I've a .doc .fileids and .transcription,
    but please tell me if I'm right, in the .transcription I ve this:
    ILEM (Ilem00)
    ILEM (Ilem01)
    ...
    ILEM (Ilem09)
    YEN (Yen00)
    YEN (Yen01)
    .....
    YEN (Yen08)
    ...
    right?
    then I did this "Generate the CMU binary form (DMP) " and "Converting model
    into DMP format"
    then I went to "Adapting the default acoustic model", everything went well,
    but I didn't understand one step here: "Accumulating observation counts"
    "...Make sure the arguments in bw command should match the parameters in
    feat.params file inside the acoustic model folder..." I don't konw what to
    change inside feat.params

    this it, now I'm lost, I don't know if I've to go to the training "Training
    Acoustic Model For CMUSphinx" or what else...

    I'll be so thankful if you help me, and tell if I'm missing something.
    thank you

     
  • Nickolay V. Shmyrev

    then I went to "Adapting the default acoustic model", everything went well,
    but I didn't understand one step here:

    So did you get the adapted model or not? Were there errors during the
    adaptation? Did you test the adapted model as described in tutorial? What
    accuracy did you get?

    I suspect you didn't finish the adaptation properly in that case you might
    want to revisit it and ask a questions about it.

     
  • med

    med - 2012-03-07

    TOTAL Percent correct: 50.00% Error: 50.00% Accuracy= 50.00%
    but I this was just a test and I just used some data because I don't know how
    to add move than one .wav file for the same word...,I think I don't have a
    problem here except what I said before about "...Make sure the arguments in bw
    command should match the parameters ..." I didn't change anything here...
    please tell me about .dic, how to write in it if there is two or three waysto
    pronounce a word..
    and should I go for training now?
    and last question, can I creat a java application based on pocketsphinx or
    I've to use sphinx4 for that?
    I'm sorry I know.. too many questions :s

     
  • iabdelgawad

    iabdelgawad - 2012-03-08

    Salmo 3alekom

    Thanks for this system i'm new member in this site and i need your help in our
    Graduation Project
    That we will do an application that used Arabisc to arabic voice recognition
    it will be mobile based
    - Android as the Os and it will convert Arabic voice to text and use some feature words to do some
    actions it will be similar to Edwin in Android Market And Voice Actions in
    Google we want connect to
    The Arabisc Engine to recognize and covert voice to text without using Google
    Engine in recognize (VR)

    if you have a time to help me i have some questions :

    1-How does Arabisic Work (easy tutorial)?
    2-How can Processed Arabisc on mobile environment ?

    i need the answer quickly for necessary....Thanks !

     
  • med

    med - 2012-03-10

    No answer?!!! :s :/

     
  • Nickolay V. Shmyrev

    I don't know how to add move than one .wav file for the same word...,

    You can give them different names like

    word_1.wav
    word_2.wav
    

    TOTAL Percent correct: 50.00% Error: 50.00% Accuracy= 50.00%

    I think you still have to work more on the setup. You are welcome to share all
    your files you are using in a single folder in order to get help. You can pack
    them into archive and upload to dropbox. Then give a public link here.

    please tell me about .dic, how to write in it if there is two or three
    waysto pronounce a word..

    Dictionary can support multiple pronunciation. See cmudict for details

    and should I go for training now?

    If you get enough accuracy with adaptation there is no need to train.
    Adaptation creates an accurate model

    can I creat a java application based on pocketsphinx or I've to use sphinx4
    for that?

    If you are looking for desktop application you can try sphinx4. On Android
    there is a pocketsphinx example.

     
  • Nickolay V. Shmyrev

    1-How does Arabisic Work (easy tutorial)?

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

    2-How can Processed Arabisc on mobile environment ?

    It's better to train your own semi-continuous models for the optimal mobile
    performance

     
  • emytone

    emytone - 2012-04-09

    @medlag
    you say you have accuracy of 50% , how the adaptation work on arabic , how if
    you say
    واحد
    how to search on it it is not on dictionary , kindly reply , i also try to
    build model for arabic so i go to build model from scratch training

     
1 2 > >> (Page 1 of 2)

Log in to post a comment.