gooofy / zamia-speech

Open tools and data for cloudless automatic speech recognition
GNU Lesser General Public License v3.0
443 stars 84 forks source link

./run-adaptation.sh : Lots of files missing errors during the process #41

Closed cogmeta closed 5 years ago

cogmeta commented 5 years ago

During the ./run-adaptation.sh process following errors are raised. I have been manually copying these files from other models (but I assume that is certainly wrong). Why are these files missing? Any way to fix this.

data/lang.adapt_test/phones/silence.csl data/lang.adapt_test/phones/disambig.in data/lang.adapt_test/L_disambig.fst

cogmeta commented 5 years ago

Also, i get following error which i got rid of removing offending entries.

Checking data/local/dict/lexicon.txt --> reading data/local/dict/lexicon.txt --> text seems to be UTF-8 or ASCII, checking whitespaces --> text contains only allowed whitespaces --> ERROR: lexicon.txt contains word bolting” with empty pronunciation. --> ERROR: lexicon.txt contains word crater” with empty pronunciation. --> ERROR: lexicon.txt contains word célèbre with empty pronunciation. --> ERROR: lexicon.txt contains word grâce with empty pronunciation. --> ERROR: lexicon.txt contains word müller with empty pronunciation. --> ERROR: lexicon.txt contains word müller's with empty pronunciation. --> ERROR: lexicon.txt contains word être with empty pronunciation. --> ERROR: lexicon.txt contains word ” with empty pronunciation.

Checking data/local/dict/extra_questions.txt ... --> reading data/local/dict/extra_questions.txt --> text seems to be UTF-8 or ASCII, checking whitespaces --> text contains only allowed whitespaces --> data/local/dict/extra_questions.txt is OK --> ERROR validating dictionary directory data/local/dict (see detailed error messages above)

Error validating directory data/local/dict

cogmeta commented 5 years ago

I used following commands

./speech_kaldi_adapt.py data/models/kaldi-generic-en-tdnn_f-r20180901 dict-en.ipa number.jsgf cogmeta-number

cd data/dst/asr-models/kaldi/cogmeta-number

./run-adaptation.sh

cogmeta commented 5 years ago

creating G.fst... Current configuration: [NAME] [DEFLT] [VALUE] -compile no no -fsg -fsm G.fsm -help no no -jsgf G.jsgf -symtab -toprule

INFO: jsgf.c(706): Defined rule: INFO: jsgf.c(706): Defined rule: INFO: jsgf.c(706): Defined rule: PUBLIC INFO: jsgf.c(706): Defined rule: INFO: jsgf.c(706): Defined rule: INFO: jsgf.c(706): Defined rule: INFO: jsgf.c(706): Defined rule: INFO: jsgf.c(706): Defined rule: INFO: jsgf.c(706): Defined rule: INFO: jsgf.c(706): Defined rule: INFO: jsgf.c(706): Defined rule: INFO: jsgf.c(706): Defined rule: INFO: jsgf.c(706): Defined rule: INFO: main.c(118): No -toprule was given; grabbing the first public rule: '' of the grammar 'org.zamia.number'. ERROR: "jsgf.c", line 340: Undefined rule in RHS: INFO: fsg_model.c(897): Writing FSM file 'G.fsm' mkgraph.sh: expected data/lang.adapt_test/phones/silence.csl to exist

cogmeta commented 5 years ago

mkgraph.sh: expected data/lang.adapt_test/phones/disambig.int to exist

cogmeta commented 5 years ago

tree-info exp/adapt/tree tree-info exp/adapt/tree fsttablecompose data/lang.adapt_test/L_disambig.fst data/lang.adapt_test/G.fst fstpushspecial fstdeterminizestar --use-log=true fstminimizeencoded ERROR (fsttablecompose[5.5.162~1420-ca32c]:Input():kaldi-io.cc:756) Error opening input stream data/lang.adapt_test/L_disambig.\ fst

[ Stack-Trace: ] kaldi::MessageLogger::HandleMessage(kaldi::LogMessageEnvelope const&, char const) kaldi::FatalMessageLogger::~FatalMessageLogger() kaldi::Input::Input(std::__cxx11::basic_string<char, std::char_traits, std::allocator > const&, bool) fst::ReadFstKaldi(std::cxx11::basic_string<char, std::char_traits, std::allocator >) main libc_start_main _start

gooofy commented 5 years ago

I suspect there is something wrong with your number.jsgf grammar

INFO: jsgf.c(706): Defined rule: INFO: jsgf.c(706): Defined rule: INFO: jsgf.c(706): Defined rule: PUBLIC INFO: jsgf.c(706): Defined rule: INFO: jsgf.c(706): Defined rule: INFO: jsgf.c(706): Defined rule: INFO: jsgf.c(706): Defined rule: INFO: jsgf.c(706): Defined rule: INFO: jsgf.c(706): Defined rule: INFO: jsgf.c(706): Defined rule: INFO: jsgf.c(706): Defined rule: INFO: jsgf.c(706): Defined rule: INFO: jsgf.c(706): Defined rule: INFO: main.c(118): No -toprule was given; grabbing the first public rule: '' of the grammar 'org.zamia.number'. ERROR: "jsgf.c", line 340: Undefined rule in RHS: INFO: fsg_model.c(897): Writing FSM file 'G.fsm'

cogmeta commented 5 years ago

Ok, but i had same exact errors for the "control" JSGF grammar and custom language model. And again similar set of errors with kaldi-adapt-lm python package.

cogmeta commented 5 years ago

ok. I figured it out. @gooofy This is an awesome work. Thank you so much. closing it.

gormonn commented 4 years ago

ok. I figured it out. @gooofy This is an awesome work. Thank you so much. closing it.

i have similar error. Can you tell more?