0

I'm working on a project where a user can say a word and my program will navigate to the first instance of it in an open file.

There are sometimes some words in the file that aren't in Sphinx's dictionary though. I'm hoping that I can scan through the file on startup and for each word, ask Sphinx if it knows of that word. If not, I'll ask the user to pronounce it, and then somehow add the word and the pronunciation into Sphinx.

I know PocketSphinx has a ps_add_word function which can do something like this, but I don't see anything similar for Sphinx.

Also, is there a way to train an existing word at runtime? Let's say that whenever I say "capital" Sphinx thinks I'm saying "cattle". Is there a way to train Sphinx to recognize "capital" better at runtime? I imagine it will be similar to adding a new word.

I'm excited at the possibility to move off of Dragon to Sphinx, so any help will be greatly appreciated. Thanks!

Verdagon
  • 2,456
  • 3
  • 22
  • 36
  • Unfortunately this feature is not supported. – Nikolay Shmyrev Oct 12 '15 at 07:10
  • @Nikolay I also am doing something similar to what Verdagon explains here. My idea was to pause the recognition system and manually build the resources, and re initiate the recognizer with the them. What is your opinion on doing something like that? – Ahmed Shariff Nov 01 '15 at 06:41
  • 1
    @AhmedShariff, I think there should be no problem with that, it should work. The problem is to make it in runtime which is not supported. If you already collected the data you can update the model with simple scripts while engine is paused and restart the engine. – Nikolay Shmyrev Nov 01 '15 at 19:55
  • @nikolay thank you, and yeah, for now i am only considering the language model and maybe the dictionary, i think it's best save working with acoustic models for later. i am not sure if this is the place to ask this, has a work around been made for being able to switch models in live speech recognition? – Ahmed Shariff Nov 02 '15 at 08:14

0 Answers0