Changes between Version 4 and Version 5 of AudioSegmentation


Ignore:
Timestamp:
12/07/06 20:03:51 (15 years ago)
Author:
anonymous
Comment:

--

Legend:

Unmodified
Added
Removed
Modified
  • AudioSegmentation

    v4 v5  
    1414 
    1515{{{ 
    16 Date :   Wed, 6 Dec 2006 21:27:29 +0800 
    17 De :    "Xie Zhiqing" <kramxxx@gmail.com> 
    18 À :     htk-users@eng.cam.ac.uk 
    19 Objet:  [HTK-Users] Lightly supervised acoustic training 
    20  
    21 Hi, my name is Mark and I am a student from Singapore.  Currently I am 
    22 working on a project on speech recognition, specifically on trainning 
    23 portion of the system.  From my knowledge, basically what I am 
    24 supposed to do is to train the system on a small amount of manually 
     16train the system on a small amount of manually 
    2517transcribed speech (.wav and .lab) and then use it to transcribe a 
    2618larger amount of untranscribe speech (only with .wav).  If the 
    27 confidence level is high enough, then it will be added into the 
     19confidence level is high enough, then add into the 
    2820trainning data and the process will be run iteratively until all the 
    29 untranscribe speech is added to the trainning data.  Is this method 
    30 correct? 
    31  
    32 >From what I gather from the HTK book, using Hvite it will output 
    33 transcriptions for the raw speech.  The format is as such => start 
    34 time , end time , phoneme and the total log probability.  Is there a 
    35 connection between the confidence measure and the total log 
    36 probability?  I am currently using Matlab to implement HTK. 
    37  
    38 Are there any good sites that can explain the process in more layman  
    39 terms? 
     21untranscribe speech is added to the trainning data. 
    4022 
    4123}}}