0

I'm working on pocket-sphinx android demo .. I have tried my model on the PC .. but on the mobile it is not as good as the PC .. I thought of modifing the code to insert an audio file instead of recording from the mic of the mobile . Can anyone tellme what can I change in the code to do that? Thanks in advance

Nour Reda
  • 1

1 Answers1

0

You need to dump audio files and listen for them and try to recognize them. If there are breaks it means your recognizer is not fast enough, you need to optimize the model size.

Nikolay Shmyrev
  • 24,897
  • 5
  • 43
  • 87
  • I still cant modify the code to take audio file instead of the mic and upon using the mic the app doesn't respond and give random words. – Nour Reda Apr 19 '19 at 20:22
  • You do not need to modify the code, you dump the files and recognize them on PC, you can also simply listen to them. – Nikolay Shmyrev Apr 20 '19 at 21:02
  • I did that but the raw files are either empty or noise although it detects, could it be that i'm using large lm and not gram the model works fine on PC??? – Nour Reda Apr 23 '19 at 21:20
  • Share them. Yes, big model will not work as described at https://stackoverflow.com/questions/25949295/cmusphinx-pocketsphinx-recognize-all-or-large-amount-of-words/25951895#25951895 – Nikolay Shmyrev Apr 24 '19 at 08:12
  • those are the raw files https://drive.google.com/open?id=1a_I_tWmt-IXFvgBzgK8sJ5tN7_zxEILk – Nour Reda Apr 25 '19 at 12:46
  • The audio is ok, you can listen in audacity.Now how do you configure the model and also show the logcat output. – Nikolay Shmyrev Apr 25 '19 at 12:55
  • Also can i change the initial number of densities when i change it i get an error that the mean file is defined as as initial 1 and stops before training – Nour Reda Apr 26 '19 at 14:02