is training the acoustic model per sentence is wrong ?

[Bryan]

sir if i train like this

   <s> hello good morning </s> (sound1)

the sound1 contains the word hello good morning separated by pause
is traning like this is good or bad
or should i train per word not in sentence?

[Nickolay V. Shmyrev]

is traning like this is good or bad

It is ok. If you want to make training more accurate you can insert <sil> words in your transcription to make trainer insert silence between words:

   <s> hello <sil> good morning </s> (sound1)

or should i train per word not in sentence?

You should train sentences, not words. If you have questions like this one you can just read a tutorial

http://cmusphinx.sourceforge.net/wiki/tutorialam

All the questions are answered there

[FF]

Hi Nickolay, I have a similar problem.
I want to move a robot with command like <command> <some words for directions>.
I want to train my system to be the most generic possibile, so everyone move my robot.
I should train with sentences like

<s> go forward</s> sound1
<s> come back</s> sound2

or is it better like <s>go forward <silence> come back </s> sound1
with some commands together?
I'd like to put sentences like "go forward" "go straight forward" to make it more friendly to users.
What do you suggest me ?

[Nickolay V. Shmyrev]

What is your guess on that problem?