Menu

Failed to align audio to trancript Error

Help
toneemy
2012-06-27
2012-09-22
  • toneemy

    toneemy - 2012-06-27

    I have this error while training my Aquestic model , in phase 3 forward-
    backward

      0   316 6 ERROR: "backward.c", line 430: Failed to align audio to trancript: final state of the search is not reached
    ERROR: "baum_welch.c", line 331: Shereen_Fawzy/Shereen_4 ignored
    

    please , i need fast reply
    thanks in advance

     
  • toneemy

    toneemy - 2012-06-27

    to solve this i remove all files cause this error to happen
    but finally i get this error

    lda.py failed to create LDA transform with status 0

    what to do now , please any help
    thanks in advance

     
  • Nickolay V. Shmyrev

    Please check lda log in logdir for details

    Please make sure that you followed the instructions to install dependencies
    precisely:

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

     
  • toneemy

    toneemy - 2012-06-27

    thanks
    please may i ask some questions
    1) what is the diffrence between semi and cont model , how to decide which i
    need , as i use pocket sphinx and is there specific configuration for feat

    2) if i build my aquestic model can i use it with exepanded language model and
    dictionary with new vocabulary have the same phones i train my aquestic with
    it.
    3)when i try to build my language model i fell interuption , what sntence to
    use , and how much the words occures, as i build small model for controlling
    mobile but while i build small lm for tidigides i t recognize the numbers that
    come togrther faster than another , so please tell me how to build good lm
    4) ldamlt not important for pocketsphinx , as i read here http://cmusphinx.so
    urceforge.net/wiki/ldamllt
    .
    sorry for this questions but i need it very much.
    thanks in advance :)

     
  • Nickolay V. Shmyrev

    1) what is the diffrence between semi and cont model , how to decide which i
    need , as i use pocket sphinx and is there specific configuration for feat

    Semi-continuous uses different scoring method for scoring the senones. In
    semi-continuous model gaussians are shared across all senones and in
    continuous each senones have it's own gaussians. Semi-continuous models are
    faster but less accurate, continuous are slower but more precise. If you need
    speed, choose semi-continuous, if you need accuracy, continuous. If you need
    compatibility with sphinx4, choose continuous because sphinx4 doesn't support
    semi-continuous models.

    2) if i build my aquestic model can i use it with exepanded language model
    and dictionary with new vocabulary have the same phones i train my aquestic
    with it.

    Yes

    3)when i try to build my language model i fell interuption , what sntence to
    use , and how much the words occures, as i build small model for controlling
    mobile but while i build small lm for tidigides i t recognize the numbers that
    come togrther faster than another , so please tell me how to build good lm

    This issue is covered by tutorial

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

    4) ldamlt not important for pocketsphinx , as i read here

    You misunderstood the page. LDA doesn't apply for semi-continuous models, it's
    not about pocketsphinx.

     
  • toneemy

    toneemy - 2012-06-27

    a lot of thanks.
    another quetion please
    1) while building my model i have this error in MODULE: 45 Prune Trees
    FATAL: "main.c", line 167: Unable to open
    /home/emytone/sphinx/fa4e/trees/fa4e.unpruned/ث-0.dtree for reading; No such
    file or directory
    what to do here?
    2) also in MODULE: 50 Training Context dependent models

    INFO: feat.c(684): Initializing feature stream to type: '1s_c_d_dd',
    ceplen=13, CMN='current', VARNORM='no', AGC='none'
    INFO: cmn.c(142): mean= 12.00, mean= 0.0
    INFO: main.c(283): Reading
    /home/emytone/sphinx/fa4e/model_architecture/fa4e.1000.mdef
    WARN: "model_def_io.c", line 436: Unable to open
    /home/emytone/sphinx/fa4e/model_architecture/fa4e.1000.mdef for reading; No
    such file or directory
    FATAL_ERROR: "main.c", line 1905: initialization failed
    thanks in advance

     
  • Nickolay V. Shmyrev

    FATAL: "main.c", line 167: Unable to open
    /home/emytone/sphinx/fa4e/trees/fa4e.unpruned/ث-0.dtree for reading; No such
    file or directory what to do here?

    The phone symbols must be alphanumeric, UTF-8 symbols are not allowed. See for
    details

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

    2) also in MODULE: 50 Training Context dependent models

    This error is caused by the earlier errors.

     
  • toneemy

    toneemy - 2012-06-28

    well ,

    ||The phone symbols must be alphanumeric, UTF-8 symbols are not allowed. See
    for details http://cmusphinx.sourceforge.net/wiki/tutorialam

    i read this tutorial many times , but no information about UTF-8 , and i
    already built small model for this chars
    ء ب ت ح خ د ر ز س ش ص ط ع ل م ن ه و ي ِ ُ

    and now while tring to build new model with the remaining phones in the arabic
    language , the trainer refuse to recognize the new phones which is (ث ج ذ ش ض
    ظ غ ف ق ك )
    as you will note in the first model it accepts the phone ش but in the second
    it does not accept , what is the reason , and how to gab this , is this mean i
    can not build my arabic language model ,
    Note : my first model was semi, but the second is cont, is this may affect the
    result

    ,
    2) i try to replace the phones it does not support by the phones it accept ,
    it does not right , i know , but as tring to see how it will recognize with
    that but i face the problem of

    ERROR: "gauden.c", line 1667: Variance (mgau= 99, feat= 0, density=0, component=21) is less then 0. Most probably the number of senones is too high for such a small training database. Use smaller $CFG_N_TIED_STATES.
    

    but i small it from 1000 to 100 , and also does not work still the error ,
    while in th efirst model my vocab is less than now and it run on 200 senones ,
    what to do , i fell I am not understand any thing , please forgive me for many
    questions , but , realy i do not know what to do

     
  • toneemy

    toneemy - 2012-06-28

    please , @nshmyrev , answer me please , what to do in the problem i state in
    th eprevioes reply .

     
  • Nickolay V. Shmyrev

    read this tutorial many times , but no information about UTF-8 , and
    i already built small model for this chars ء ب ت ح خ د ر ز س ش ص
    ط ع ل م ن ه و ي ِ ُ

    There is information. It says only alphanumeric. UTF-8 is not included. Please
    read the text accurately.

    is this mean i can not build my arabic language model ,

    Use alphanumeric phone names

    i fell I am not understand any thing , please forgive me for many questions
    , but , realy i do not know what to do

    The first thing you need to do is to provide more information. It's hard to
    say what exactly is the reason until that. For more details see the
    corresponding
    tutorial section on troubleshooting

    http://cmusphinx.sourceforge.net/wiki/tutorialam#troubleshooting

    One of the certain things to do is to add more data to your training set.
    Please
    read the tutorial about the amount of data required:

    http://cmusphinx.sourceforge.net/wiki/tutorialam#when_you_need_to_train

     

Log in to post a comment.