[PD] speech recognition for pd

[patrick]

hi all,

last night, i did a bridge between sphinx3 and pd. sphinx3 is an open 
source speech recognition software.

have a look at the demo
http://www.vimeo.com/2814913

i am looking for help for optimizing or even better making an external 
for it.

1) working on linux only for now.

2) using shell to call a python script (py/ext is crashing when import 
_sphinx3)

3) communication between the python script and pd via open sound control 
(mrpeach) - i like this solution, cause i can use the same script to 
have speech recognition in blender or other osc compatible software

4) using shell again to call sox (for converting the soundfile to raw). 
is it possible to write a raw soundfile in pd (1 channel, signed, 16000 
(samplerate))?

5) dynamic patching with send pd-$0.sphinx (obj 20 20 $voicerecognition) 
- would like to be able to connect objects together, need to learn more 
about dynamic patching

6) it's possible to make a custom dictionary, for now i will create one 
for basic stuff (http://wiki.puredata.info/en/Category:vanilla)

7) tested with two big accents : french (me) and japanese (my girlfriend)

pat