use the latest grapheme to phonem generated by g2p_train(sequitur)

local/g2p.sh

@@ -25,7 +25,7 @@ out_lexicon=$3
 
 [ ! -f $vocab ] && echo "Can't find the G2P input file: $vocab" && exit 1;
 
-sequitur_model=$g2p_model_dir/model-3
+sequitur_model=$g2p_model_dir/model-5
 
 # Turns out, that Sequitur has some sort of bug so it doesn't output pronunciations
 # for some (admittedly peculiar) words. We manually specify these exceptions below