Custom Model created by me has a small size

Hi ,

I wanted to use the pre trained output_graph.pb file for detecting my voice for 6 words .(left ,right,up,down ,on,off) .
I created a new vocab.txt file having the sentences for these 6 words . After this I created the words.arpa , lm.binary and trie file , but the size of files created by me is very less(lm.binary =125MB,output_graph.pb =16MB) and it is not able to recognize any of the words.
I have followed the steps mentioned in the tutorial-how-i-trained-a-specific-french-model-to-control-my-robot
Please help me out if I am missing something.
Also I have observed that the model is able to recognize those .wav files whose amplitued is between 0.5 - (-0.5) . If there any way to increase this.

This is not consistent with what you say earlier: use the pre trained output_graph.pb, you should have a file ~180MB.