Weird pocketsphinx accuracy degredation

Question

A little bit of context - I'm creating a pocketsphinx application for android, using jsgf grammars (though the problem persists with fsg grammars as well). My code is based on the pocketsphinx demo for android.

Pocketsphinx is producing a bizarre issue where the accuracy appears to be degrading after only a couple of queries to the engine. The first query has near perfect accuracy - it can recognize relatively complicated and convoluted phrases without difficulty. However the second, and third recognitions can barely pick up a two syllable word, and by the fourth query to the engine, it simply fails to generate a hypothesis.

I'm not quite sure whats causing the problem, the original demo appeared to work fine, and I have hardly altered the configuration of the speech engine, other than adding a grammar. Anyway, has anyone else ever experienced something like this, or even have a suggestion as to what I could try to remedy this?

Thanks

It trains at runtime? Is there an option to pass to the decoder to disable this? Haven't used pocketsphinx all that long, I thought training occurred only through the use of tools during development. — lafuzz, Aug 08 '12 at 00:07
You already asked here http://sourceforge.net/projects/cmusphinx/forums/forum/5471/topic/5507338 — Nikolay Shmyrev, Aug 08 '12 at 20:05
Wasn't aware it was bad practice to ask questions on totally different sites... I'll refrain from that in the future. — lafuzz, Aug 08 '12 at 21:33

score 1 · Answer 1 · answered Aug 08 '12 at 21:34

1

Problem solved, turns out I needed the -dither option to account for 'zero energy regions' Thanks to Nikolay Shmyrev for the solution https://sourceforge.net/projects/cmusphinx/forums/forum/5471/topic/5507338

answered Aug 08 '12 at 21:34

lafuzz

74
5

Weird pocketsphinx accuracy degredation

1 Answers1

Linked