Menu ▾ ▴

To transcribe lectures( Eng-Indian Accent)

Sphinx4 Help
vijay k
2010-08-12
2018-09-13

  • vijay k

    vijay k - 2010-08-12

    Hi all..,

    I am working on automatically transcribing lectures which are in
    english(indian accent). Since the sphinx is the best opensource speech
    recognition I have decided to use it. But I am new to sphinx 4. please any one
    help how to proceed..what are the requirements for it ? how to improve
    accuracy?

    thanks in advance.

     

  • Nasir Hussain

    Nasir Hussain - 2010-08-14

    Hey Vijay,

    I am working on automatically transcribing lectures which are in
    english(indian accent)

    For that u have create an acoustic model in your voice(indian Accent).

    Since the sphinx is the best open source speech recognition I have decided
    to use

    You are absolutely right my Friend...:)

    But I am new to sphinx 4. please any one help how to proceed

    Just go through the Site Below.It contains all the neccesary details for
    installation and Usage.
    http://cmusphinx.sourceforge.net/sphinx4/

    https://sourceforge.net/projects/cmusphinx/files/
    Download latest Sphinx4 which is Beta 4 Source file and then configure it with
    any IDE as per ur needs(like eclipse)
    then run the demos in it...And check out transcriber demo ;)

     

  • vijay k

    vijay k - 2010-08-14

    Thanks for your reply.
    can you please tell me the steps to create acoustic model?

     

  • vijay k

    vijay k - 2010-08-25

    hi nasir,

    can you please answer my following questions.

    how much time it will take to create acoustic models? and what is the
    complexity level?

    what is the accuracy level?and how to improve accuracy?

    what about dictionary and language models? new ones need to be created or
    existing ones are enough?

    how much training data is required and manually transcribed?

    can it handle out of vocabulary words?

    thanks in advance

    -vijay

     

  • Nasir Hussain

    Nasir Hussain - 2010-08-25

    Hey Vijay,

    Although Such questions can be answered Accurately By Nickolay....
    He is awesome and he helps :)
    Anyways....

    Can you please answer my following questions.

    Ya Sure...:)

    how much time it will take to create acoustic models? and what is the
    complexity level?

    Acoustic Model creation depends upon how much Sound files u Using....for
    example I created a Acoustic model of 60 UK english words and it took me for
    about 30 seconds.....If u have alot of words than it will take time
    accordingly...:)
    And its Not that complex...:P

    what is the accuracy level?and how to improve accuracy?

    Hmm...Accuracy Level is Gud....and u can increase ur accuracy By changing ur
    config File settings.... If u need My settings Just let me know...:)

    what about dictionary and language models? new ones need to be created or
    existing ones are enough?

    See U need a dictionary while creation of Acoustic model...so after creation
    of the model use that dictionary in ur app...:)...
    and language models can easily be created by just providing a corpus file here
    http://www.speech.cs.cmu.edu/tools/lmtool-
    new.html     (Remember
    language model created on the mentioned link is a US English Language
    Model)....Inorder to create ur Own Language model u have to create it using
    cmuclmtk tool...:)

    how much training data is required and manually transcribed?

    All depends upon ur needs....It can be 60 words to 60000...:)

    can it handle out of vocabulary words?

    Hell Ya...:P

    -Nasir aka Paul

     

  • vijay k

    vijay k - 2010-08-26

    thanks alot for the help.

    how to create the dictionary for indian english? do we need to write phoneme
    sequences for each word manually?

     

  • Nasir Hussain

    Nasir Hussain - 2010-08-26

    Hey Vijay,

    how to create the dictionary for indian english?

    I would recommend u to create It using lmtools..which i recommended u
    earlier.. See the concept of Indian English comes With the acoustic
    model...when u record ur voices and create the acoustic model the accent gets
    recorded....:)..just Try to use more words in ur acoustic model to get gud
    accuracy...:)

    do we need to write phoneme sequences for each word manually?

    Ya in Some cases we need to...but not in all...the pronunciation tht we get
    using lmtool is sufficient according to me....
    But Some need changing...as u and i both know Indian enlish :P

    -Nasir

     

  • sriram

    sriram - 2015-09-25

    Hi Vijay and Nasir,

    Do you have any woking database for training the Indian English Accent.
    Please share if you have any.

    Thanks and Regards
    Sriram.

     

  • akshat dashore

    akshat dashore - 2018-09-13

    hi all,
    i am a beginner and searching for indian english and hindi pretrained models for speech to text kindly share it to me if available else which please tell me how could i proceed and also i used this one : https://sourceforge.net/projects/cmusphinx/files/Acoustic%20and%20Language%20Models/Indian%20English/cmusphinx-en-in-5.2.tar.gz/download

    but it wouldn't works at all
    Thanks alot in advance
    waiting for your valuable reply

     

    • sheharyar masood

      sheharyar masood - 2018-09-13

      hi akshat,
      the link is good from where you have downloaded cmusphinx.... here is the link for your further knowledge....
      https://cmusphinx.github.io/wiki/tutoriallm/

       

Log in to post a comment.