I created an artificial arpa languagemodel, (because our use case has no real corpus, so I use some patterns to get the possible ngrams. and because we don't assume anything, so we set most of the ngrams the same probability and backoff weights).
With the FST generated fro this artifical arpa languagmodel, some of the utterances failed to be decoded with
{{{
"Epsilon loops exist in your decoding" which comes from:KALDI_ASSERT(loop_count < max_loop && "Epsilon loops exist in your decoding "
"graph (this is not allowed!)");
}}}
in lattice-faster-decoder.cc
I tried to enlarge the max_loop to be 10million, still get this error.
Could you please give me some hints as what could be the cause of this error?
Dear Yajie
I created an artificial arpa languagemodel, (because our use case has no real corpus, so I use some patterns to get the possible ngrams. and because we don't assume anything, so we set most of the ngrams the same probability and backoff weights).
With the FST generated fro this artifical arpa languagmodel, some of the utterances failed to be decoded with {{{ "Epsilon loops exist in your decoding" which comes from:KALDI_ASSERT(loop_count < max_loop && "Epsilon loops exist in your decoding " "graph (this is not allowed!)"); }}} in lattice-faster-decoder.cc
I tried to enlarge the max_loop to be 10million, still get this error.
Could you please give me some hints as what could be the cause of this error?
Best