General Discussion

Flat
VoxForge for Arduino Recognizer
User: Gerald Friedland
Date: 1/2/2017 4:41 pm
Views: 4881
Rating: 0

Hi all,

I am one of the co-developers of MOVI (http://www.audeme.com), an offline speech recognition board that integrates with Arduino. The board internally uses PocketSphinx and we have managed to package up the Spanish and German models from VoxForge in a way that we can now also use them for the recognizer (on top of English).

FYI. I wrote two Instructables on it:

http://www.instructables.com/id/Connection-less-German-Speech-Recognition-and-Synt/ 

http://www.instructables.com/id/Connection-Less-Spanish-Speech-Recognition-and-Syn/

However, I have two major questions:

a) Why are the Spanish and the German models loading and recognizing so much more slowly than the English models (that come with Pocketsphinx). In other words: Is there any optimization that I can do?

b) Does VoxForge have any other languages available as PocketSphinx models with the corresponding Phonetisaurus dictionary? (That's ultimately what we need)

Thank you for input.

Gerald

--- (Edited on 1/2/2017 4:41 pm [GMT-0600] by Visitor) ---

Re: VoxForge for Arduino Recognizer
User: nsh
Date: 1/2/2017 5:39 pm
Views: 2266
Rating: 1

> a) Why are the Spanish and the German models loading and recognizing so much more slowly than the English models (that come with Pocketsphinx). In other words: Is there any optimization that I can do?

You are using very old continuous models. You need PTM model, explained in tutorial http://cmusphinx.sourceforge.net/wiki/acousticmodeltypes
Recent PTM models are available here:
Spanish cmusphinx-es-5.2.tar.gz
German cmusphinx-ptm-voxforge-de-r20161117.tar.xz

--- (Edited on 1/3/2017 02:39 [GMT+0300] by nsh) ---

PreviousNext