Converting bin to arpa only extract unigram

[BananaNosh]

When using sphinx_lm_convert -i model.lm.bin -ifmt bin -o model.lm.arpa -ofmt arpa I get the following error and only the unigrams are extracted to the new format:

sphinx_lm_convert: ../../../../src/libsphinxbase/lm/ngram_model_trie.c:279: ngram_model_trie_write_arpa: Assertion "raw_ngram_idx == base->n_counts[i - 1]" failed.

This error is the same as in Issue #52.

I should note that the binary file was built with the tool itself from a valid arpa file. The translation back, however, then drops the bi- and trigrams with the error mentioned above. Uncommenting the assertion leads to a segmentation fault.

Cheers