ERROR: "ngram_search.c", line 1207: Failed to find DAG node corresponding to </s>

[sudopsi4]

I am getting a sentence error of 3.9% and word error rate of 4.0% with this error. what should I do?
Is this error rate good already? or is it just due to the halting of training from this error?

MODULE: DECODE Decoding using models previously trained Decoding 2362 segments starting at 0 (part 1 of 1) 0% ERROR: This step had 1 ERROR messages and 0 WARNING messages. Please check the log file for details. Aligning results to find error rate SENTENCE ERROR: 3.9% (92/2362) WORD ERROR RATE: 4.0% (94/2362)

My Log file

• INFO: ngram_search_fwdflat.c(960): fwdflat 0.00 CPU 0.001 xRT
• INFO: ngram_search_fwdflat.c(963): fwdflat 0.00 wall 0.001 xRT
• INFO: ngram_search.c(1197): not found in last frame, using .78 instead
• ERROR: "ngram_search.c", line 1207: Failed to find DAG node corresponding to
• INFO: batch.c(761): b83c1acf/5_b83c1acf_nohash_2: 0.99 seconds speech, 0.01 seconds CPU, 0.01 seconds wall
• INFO: batch.c(763): b83c1acf/5_b83c1acf_nohash_2: 0.01 xRT (CPU), 0.01 xRT (elapsed)
• (b83c1acf/5_b83c1acf_nohash_2 -1032)
• b83c1acf/5_b83c1acf_nohash_2 done

[dhdaines]

It could be a good error rate, depending on the complexity of your task (it appears to be pretty simple). It is hard to say a priori what is causing the error but it could simply be that the sentence in question is not well recognized by the language model. You can try:

• deactivating bestpath search (-bestpath no)
• widening the decoder beams (-beam 1e-80 -wbeam 1e-60 -pbeam 1e-60 for instance)

It is unlikely to be a bug in PocketSphinx but there isn't really any way to know from the information available.