Author Topic: Version 0.960  (Read 5767 times)

0 Members and 2 Guests are viewing this topic.

jitterjames

  • Administrator
  • Hero Member
  • *****
  • Posts: 7714
  • Karma: 116
    • View Profile
    • VoxCommando
Version 0.960
« on: July 30, 2012, 08:59:08 PM »
This version should address some issues with reading .wav files generated by the android app. "VoxWav", as it is currently being called.

I have also added support for SOX (Sound eXchange) in case you want to preprocess your wav files before VoxCommando analyses them. http://sox.sourceforge.net/

If you want to try using sox to do this then you need to do a few things.

Download Sox version 14.4 (or later but I've been testing with 14.4) : http://sourceforge.net/projects/sox/files/sox/14.4.0/
open the zip and put all the files directly into a subfolder of VoxCommando called sox.

example:  C:\voxcommando\sox

now open options in VoxCommando.  On the advanced tab, next to the wavwatch folder, you can now select the "Sox preprocessing" checkmart and enter the sox parameters that you want to use:

here's an example: 
Code: [Select]
fade .4 reverse fade .6 reverse vol 0.8
this will fade the sound in and out at the end (because I get a slight clicking at the beginning and end of my .wavs) and then reduce the volume to 80% of original.

EDIT: I attached 0.960b which keeps the sox window hidden during processing.  This is important if you are using XMBC fullscreen (for example) since the sox window would popup and XBMC would go out of fullscreen mode!
« Last Edit: February 04, 2015, 09:47:48 AM by jitterjames »

claymic

  • $upporter
  • Sr. Member
  • *****
  • Posts: 152
  • Karma: 0
    • View Profile
Re: Version 0.960
« Reply #1 on: July 30, 2012, 09:20:07 PM »
Man, do i already tell you how much i am your fan ? Great! Great :clap

jitterjames

  • Administrator
  • Hero Member
  • *****
  • Posts: 7714
  • Karma: 116
    • View Profile
    • VoxCommando
Re: Version 0.960
« Reply #2 on: July 30, 2012, 10:03:36 PM »
it looks like I need to hide the SoX window :-[

ok fixed.  I attached 0.960b above
« Last Edit: July 30, 2012, 10:22:58 PM by jitterjames »

claymic

  • $upporter
  • Sr. Member
  • *****
  • Posts: 152
  • Karma: 0
    • View Profile
Re: Version 0.960
« Reply #3 on: July 30, 2012, 11:30:52 PM »
Tested, tested and tested.
Works in all my devices without any problem, better than the microphone. Perfect.
Thanks Sor James.
« Last Edit: July 31, 2012, 06:59:43 AM by claymic »

Kalle

  • $upporter
  • Hero Member
  • *****
  • Posts: 2320
  • Karma: 47
    • View Profile
Re: Version 0.960
« Reply #4 on: July 31, 2012, 05:33:21 AM »
 :clap :hugs :yay ::bow ::bow ::bow ::bow

GREAT JOB
***********  get excited and make things  **********

DHHJ

  • $upporter
  • Contributor
  • *****
  • Posts: 62
  • Karma: 1
    • View Profile
Re: Version 0.960
« Reply #5 on: July 31, 2012, 11:16:47 PM »
I'm having low volume issues with VoxWav with both a Galaxy S2 and a XOOM tablet, both running ICS. The .wav file sounds fine when I play it back. But recognition is in the 40% range. I'm currently running 960b and 1.2b but have had volume issues throughout all releases. Do I need to create a new profile or use some gain with Sox?

Kalle

  • $upporter
  • Hero Member
  • *****
  • Posts: 2320
  • Karma: 47
    • View Profile
Re: Version 0.960
« Reply #6 on: August 01, 2012, 12:11:47 AM »
I'm having low volume issues with VoxWav with both a Galaxy S2 and a XOOM tablet, both running ICS. The .wav file sounds fine when I play it back. But recognition is in the 40% range. I'm currently running 960b and 1.2b but have had volume issues throughout all releases. Do I need to create a new profile or use some gain with Sox?
Test it without sox. You can also create a new untrained speech profil in windows, perhaps it helps.
« Last Edit: August 01, 2012, 12:13:56 AM by Kalle »
***********  get excited and make things  **********

jitterjames

  • Administrator
  • Hero Member
  • *****
  • Posts: 7714
  • Karma: 116
    • View Profile
    • VoxCommando
Re: Version 0.960
« Reply #7 on: August 01, 2012, 12:46:03 AM »
I'm having low volume issues with VoxWav with both a Galaxy S2 and a XOOM tablet, both running ICS. The .wav file sounds fine when I play it back. But recognition is in the 40% range. I'm currently running 960b and 1.2b but have had volume issues throughout all releases. Do I need to create a new profile or use some gain with Sox?

I would try a new profile, you can always go back to the old one if it doesn't work out.

I'm curious though, how do you know you are suffering from low volume?  VC usually works well with low volume.  And you say it "sounds fine" so maybe it is not the issue. Then again, if you trained your old profile at high volume, then maybe it is...

Boosting the volume with sox is worth a shot too.  I think you just need to write. Something like:
Code: [Select]
vol 1.5 to increase it by 50%.  Just remember that you'll be boosting all the noise too.

DHHJ

  • $upporter
  • Contributor
  • *****
  • Posts: 62
  • Karma: 1
    • View Profile
Re: Version 0.960
« Reply #8 on: August 01, 2012, 10:01:15 PM »
Its working! Strange but I turned Sox off and I've got 90%+ recognition. Probably a Layer 8 problem  :bonk
Thanks for the suggestions Kalle and James.

claymic

  • $upporter
  • Sr. Member
  • *****
  • Posts: 152
  • Karma: 0
    • View Profile
Re: Version 0.960
« Reply #9 on: August 04, 2012, 05:17:37 PM »
James
What the diference between Fade and Vol in SOX ?
Thanks

jitterjames

  • Administrator
  • Hero Member
  • *****
  • Posts: 7714
  • Karma: 116
    • View Profile
    • VoxCommando
Re: Version 0.960
« Reply #10 on: August 04, 2012, 05:43:12 PM »
vol changes the volume of the whole .wav file from beginning to end.  Fade will fade in at the beginning from 0 volume to 100% of the origianl volume.  the parameter is how long the fade lasts in seconds.  So if you use  "fade 0.4" the volume will start at 0 and then reach normal volume at 0.4 seconds into the wav file.

the reason I use reverse (twice) is that you can only fade in at the beginning.  There is no function that allows us to fade-out at the end.  By reversing and fading in, then reversing again, you are effectively fading out.

sox has quite a good manual.  You should be able to use google translate to read it.
http://sox.sourceforge.net/sox.html

there are also noise removal functions, and high and low pass filters etc.  but I tried some tests and they only seem to reduce the accuracy of VC for me.

claymic

  • $upporter
  • Sr. Member
  • *****
  • Posts: 152
  • Karma: 0
    • View Profile
Re: Version 0.960
« Reply #11 on: August 04, 2012, 06:17:25 PM »
Very Thanks James, now i understand why Fade works better when i am using my Motorola Atrix and the Galaxy Mini.
I reed the manual.