Reply by Mark A Mandel September 9, 20042004-09-09
Please take comp.speech.users off the post-to and followup-to lists for
this topic.
Reply by ranjeet September 4, 20042004-09-04
Hi all !!

  ( FIRST OF ALL I AM THANKFULL TO JERRY, TIM, JOHAN, RANDY for clearing my 
   doubts in the Floatiing and fixed point. Well i will like to know the algo
   which jerry talked about mutilpication (how it is done)).

 Now I want to share my understanding as I am Engineering student so please let
 me know about my blunder.

   I have a audio file, (8Khz, 16bit data) Now the thing is that I have to 
   compute the energy of the audio file. Let me explain in breif.. 

  I take the 10m sec data means I have 80 samples value. Now I consider this 80 
  samples value as One frame. 

  Now As G.729 annex B says that we have to process the 240 samples for
  Computing the auto corelation coefficent of the current frame.

  Now it the G.729 annex A says that You have the 120 samples as the previous 
  speech 80 as the current and the 40 as the future.  
 
  /*----------------------------------------------------------------------*
  *      Initialize pointers to speech vector.                          *
  *                                                                     *
  *                                                                     *
  *   |--------------------|-------------|-------------|------------|   *
  *     previous speech           sf1           sf2         L_NEXT      *
  *                                                                     *
  *   <----------------  Total speech vector (240)------------------>   *
  *   <----------------  LPC analysis window (240)------------------>   *
  *   |                   <-- present frame (80)------->                *
  * old_speech            |              <-- new speech (80)-------->   *
  * p_window              |              |                              *
  *                     speech           |                              *
  *                             new_speech                              *
  *-----------------------------------------------------------------------*/

  Now The thing which i undersstand is :

  For doing this I am taking the buffer pre_source_signal of 240 :
  in which i place the values after preproceesing the 80 samples.

  it goes like this  memcpy ( pre_source_signal,
                             &pre_source_signal[40],
                              160 * sizeof(double));
                  
                     memcpy ( &pre_source_signal[160],
                              preprocessed_signal,
                              80 * sizeof(double));

  Now For me the current frame to becomes the samples value of    
  pre_source_signal[120] to pre_source_signal[199] 

  and the next 40 means pre_source_signal[200] to pre_source_signal[239]
  becomes the future frame. for the Computiing the window and the LPC analysis.

  Is this correct. ????????????  
  
  Now What will be the data for the complete silence file ????? 
  the audio file ( 8khz and the 16 bit data)..  I have read that 
  for complete silence file the value is 128 is it correct or some thing else??

  The questioon came to mind as i was not able to compute the energy of the 
  file below 15 dB ......... so i wanted to compute the energy for the audio 
  file whcih is complete silence and check wht the minimum energy i will get.

  This I is my acadamic project in which I have to design the G.729 VAD 
  I dont have some many resources to validiate my self. Except U ALL and ur
  guidence.

  So plese let me know about this. 

Thanks In Advance
Ranjeet