Thanks for this very powerful project!
I am trying to install py-nltools which in turn wants to install py-kaldi-asr:
...
Processing py-kaldi-asr-0.2.4.tar.gz
Writing /tmp/easy_install-TO1N7N/py-kaldi-asr-0.2.4/setup.cfg
Running py-kaldi-asr-0.2.4/setup.py -q bdist_egg --dist-dir /tmp/easy_install-TO1N7N/py-kaldi-asr-0.2.4/egg-dist-tmp-e5cWqN
looking for atlas library, trying pkg-config first...
Package atlas was not found in the pkg-config search path.
Perhaps you should add the directory containing `atlas.pc'
to the PKG_CONFIG_PATH environment variable
No package 'atlas' found

I have a correct kaldi installation (under Ubuntu 17.10), but there is no atlas.pc, only
/usr/lib/x86_64-linux-gnu/pkgconfig/blas-atlas.pc
/usr/lib/x86_64-linux-gnu/pkgconfig/lapack-atlas.pc

Any suggestions?

Hi Guenter,

I'm trying to get your scripts running and have a question regarding the training of the German language model. I've run

speech$ ./speech_sentences_de.py --train-punkt

which writes speech/data/dst/speech/de/punkt.pickle. And

speech$ ./speech_sentences_de.py

writes speech/data/dst/speech/de/sentences.txt. So far so good. Now I'd like to run speech_build_lm.py, but according to the lines

SOURCES = ['data/dst/speech/%s/sentences.txt',
           'data/dst/speech/%s/ai-sentences.txt']

it also needs the file 'data/dst/speech/%s/ai-sentences.txt'. The command

speech$ grep -rF ai-sentences.txt .

yielded

speech_build_lm.py:           'data/dst/speech/%s/ai-sentences.txt']

So the question is: What does ai-sentences.txt contain and how do I create it? To train the language model ai-sentences.txt is not necessary, because we have sentences.txt. But I'd like to know where ai-sentences.txt comes from 😄

Thanks for your help in advance!

Cheers,
Marc

Hello, I am training the kaldi-tuda dataset using the kaldi-run-nnet3.sh but I have problems when running train_tdnn.sh. It seems that script was deprecated. What version of kaldi are you using to traing the models you have made available in voxforge?
Since it is deprecated, they point to train_dnn.py but then I run into other problems.

When running train_tdnn.sh I am getting:
steps/nnet3/train_tdnn.sh: THIS SCRIPT IS DEPRECATED steps/nnet3/train_tdnn.sh --stage -10 --num-epochs 8 --num-jobs-initial 2 --num-jobs-final 14 --splice-indexes -4,-3,-2,-1,0,1,2,3,4 0 -2,2 0 -4,4 0 --feat-type raw --online-ivector-dir exp/nnet3/ivectors_train --cmvn-opts --norm-means=false --norm-vars=false --initial-effective-lrate 0.005 --final-effective-lrate 0.0005 --cmd run.pl --pnorm-input-dim 2000 --pnorm-output-dim 250 --minibatch-size 128 data/train data/lang exp/tri1_ali exp/nnet3/nnet_tdnn_a feat-to-dim scp:exp/nnet3/ivectors_train/ivector_online.scp - steps/nnet3/train_tdnn.sh: creating neural net configs steps/nnet3/make_tdnn_configs.py --splice-indexes -4,-3,-2,-1,0,1,2,3,4 0 -2,2 0 -4,4 0 --feat-dim 13 --ivector-dim 100 --pnorm-input-dim 2000 --pnorm-output-dim 250 --use-presoftmax-prior-scale true --num-targets 1925 exp/nnet3/nnet_tdnn_a/configs steps/nnet3/train_tdnn.sh: calling get_egs.sh steps/nnet3/get_egs.sh --cmvn-opts --norm-means=false --norm-vars=false --feat-type raw --online-ivector-dir exp/nnet3/ivectors_train --transform-dir exp/tri1_ali --left-context 10 --right-context 10 --samples-per-iter 400000 --stage 0 --cmd run.pl --frames-per-eg 8 data/train exp/tri1_ali exp/nnet3/nnet_tdnn_a/egs steps/nnet3/get_egs.sh: invalid option --feat-type

Can you give any suggestions?

hey...
i want to run zamia-speech on xilinx zcu102 board with debian
arm architecture is arm64

Introduction

Currently, most scripts offer the argument --lang to choose between de and en. The choice de means that a language model is trained on the text corpora

• Europarl
• Parole

and that an acoustic model is trained on

• a corrected version of the VoxForge corpus
• a corrected version of the TU-Darmstadt corpus (gspv2)

There is no way to choose for example just one text corpus and one audio corpus. Hereby I propose to change the command line arguments of some scripts to make it simpler to pick text and audio corpora to train an ASR system.

Current workflow

An example workflow to train a German speech recognition system might look like

$ ./speech_sentences_de.py
$ ./speech_build_lm.py --lang de
$ ./speech_kaldi_export.py --lang de
$ cd data/dst/speech/de/kaldi
$ ./build-lm.sh
$ ./run-chain.sh

Under the hood speech_sentences_de.py extracts sentences from two text corpora (Europarl and Parole) and writes them to a text file containing one sentence per line. Both of the corpora have to be parsed in a unique way to extract sentences from them. Currently, there is no way to build a language model just on exactly one corpus (except by altering the script of course).

The command speech_build_lm.py --lang de concatenates text files containing one sentence per line and builds a language model (3-gram) using the program ngram.

The command speech_kaldi_export.py --lang de consumes all transcripts for the VoxForge and TU-Darmstadt corpora in data/src/speech/de/transcripts_*.csv, the pronunciation dictionary data/src/speech/de/dict.ipa and the language model created in the previous step and "deploys" them to speech/data/dst/speech/de/kaldi in a way that adheres to the Kaldi interface (regarding directory and file structure). Currently there is no way to conveniently choose exactly one, two, or more audio corpora. But it would be nice to be able to choose a small audio corpus to do regression testing.

The script build-lm.sh converts the language model from the ARPA format to the finite state transducer G.fst - as Kaldi expects it.

Finally, run-chain.sh uses Kaldi to train the acoustic model.

Here is an example sequence of commands to train a complete English ASR system with Kaldi:

$ ./speech_sentences_en.py
$ ./speech_build_lm.py --lang en
$ ./speech_kaldi_export.py --lang en
$ cd data/dst/speech/en/kaldi
$ ./build-lm.sh
$ ./run-chain.sh

Proposed workflow

This is an example of the proposed workflow for creating an ASR system:

$ ./speech_sentences.py europarl-de
$ ./speech_sentences.py parole
$ ./speech_sentences.py voxforge-de-prompts
$ ./speech_build_lm.py europarl-de \
                       parole \
                       voxforge-de-prompts \
                       lm-europarl-de-parole-voxforge-de-prompts
$ ./speech_kaldi_export.py --audio-corpus voxforge \
                           --audio-corpus tu-darmstadt \
                           --language-model lm-europarl-de-parole-voxforge-de-prompts \
                           --dictionary dict-de.ipa \
                           --model-name experiment1
$ cd data/dst/speech/kaldi/experiment1
$ ./build-lm.sh
$ ./run-chain.sh

In the example we create an ASR system where the language model is based on the Europarl and Parole corpora. To train the acoustic model, the VoxForge and TU-Darmstadt corpuora are used.

The command ./speech_sentences.py TEXTCORPUS writes the extracted sentences by convention to data/dst/text-corpora/TEXTCORPUS.txt.

The command speech_build_lm.py TEXTCORPUS [TEXTCORPUS ...] LMNAME expects as arguments a list of text corpora and a name for the resulting language model (LMNAME). The files of the language model are written to the directory data/dst/lm/LMNAME/.

The command speech_kaldi_export.py expects one or more --audio-corpus, exactly one --language-model, and exactly one --model-name MODELNAME. The script creates a directory data/dst/asr-models/kaldi/MODELNAME and places all files required by Kaldi in it.

@gooofy What do you think about my suggested changes?

UPDATES 2018-04-03

• Augmented example workflow to show that speech_sentences.py can also extract sentences from VoxForge's prompts.
• In example for proposed workflow: add --dictionary to speech_kaldi_export.py

How do I set up installation on Mac Osx ?
Thank you

Hi.

I followed the new modular approach from README.md (for German). It worked smoothly! I have some minor edits that I can put into a pull request.

In contrast to my previous zamia-speech experiments, I did not add the sentences from the speech corpora to the language model (i.e. I only processed europarl_de and parole_de with speech_sentences.py) because I think that this is a fairer evaluation. As expected, the WER jumped up. Does the following make sense or has anybody achieved lower WERs in a comparable setup?
%WER 11.57 [ 20312 / 175507, 3225 ins, 2930 del, 14157 sub ] exp/nnet3_chain/tdnn_sp/decode_test/wer_9_1.0
%WER 13.11 [ 23010 / 175507, 3531 ins, 3211 del, 16268 sub ] exp/nnet3_chain/tdnn_250/decode_test/wer_8_1.0
I configured the ASR with:
./speech_kaldi_export.py generic-de-small dict-de.ipa generic_de_lang_model voxforge_de gspv2 forschergeist zamia_de zamia_de_noisy

Hello,

i am trying to test pre-trained Kaldi models with german language. It shows errors:

debian-admin@debian:~$ python kaldi_decode_wav.py -v -m /opt/kaldi/model/kaldi-generic-de-tdnn_sp Wie_geht_es_dir.wav
DEBUG:root:/opt/kaldi/model/kaldi-generic-de-tdnn_sp loading model...
DEBUG:root:/opt/kaldi/model/kaldi-generic-de-tdnn_sp loading model... done, took 52.890310s.
DEBUG:root:/opt/kaldi/model/kaldi-generic-de-tdnn_sp creating decoder...
DEBUG:root:/opt/kaldi/model/kaldi-generic-de-tdnn_sp creating decoder... done, took 0.695156s.
Traceback (most recent call last):
File "kaldi_decode_wav.py", line 72, in
if decoder.decode_wav_file(wavfile):
File "kaldiasr/nnet3.pyx", line 195, in kaldiasr.nnet3.KaldiNNet3OnlineDecoder.decode_wav_file (kaldiasr/nnet3.cpp:4362)
AssertionError

I had installed with this guide https://github.com/gooofy/zamia-speech#debian-9-stretch-amd64

Can you tell me, why is not working :(. Thank you.

Dear Guenter,
It will be very helpful to know hardware requirements, to avoid problems with lack of RAM, HDD or GPU RAM and wasting time to trying to train using not enough powerful computer.

I understand that there may be no well defined requirements, everything depends from used corpora, configs, etc.

But would you mind at least sharing your hardware spec to understand what was enough to build kaldi-generic-en-tdnn_f model. And how much time it took.
Thank you!

The file train_all.txt generated during the construction of the German language model contains many English sentences. Some boiler plate sentences can be removed from data/dst/text-corpora/ with a tiny grep call, but I would like to exclude whole texts from parole_de because they are multilingual, e.g. geheimdienst.sgm. My current solution is to rename such files so that their suffix is not .sgm anymore (that's what the script is looking for). Is there a cleaner approach like an exclude file?

*_to_vf.sh scripts try to write a script called tmp/run_parallel.sh, yet tmp is not created beforehand, yielding a crash.

I Downloaded kaldi-generic-en-tdnn_sp-r20190227 and tried to use it with the kaldi online decoders.
I used the following arguments:

./online2-tcp-nnet3-decode-faster --feature-type=mfcc --min-active=200 --max-active=8000 \
--beam=13.0 --lattice-beam=6.0 --acoustic-scale=1.0 --frames-per-chunk=51 \
--frame-subsampling-factor=3 --endpoint.silence-phones=1:2:3:4:5:6:7:8:9:10:11:12:13:14:15 --ivector-silence-weighting.silence-weight=1e-3 --ivector-silence-weighting.silence-phones=1:2:3:4:5:6:7:8:9:10:11:12:13:14:15 \
--ivector-extraction-config=/home/models/kaldi-generic-en-tdnn_sp-r20190227/ivectors_test_hires/conf/ivector_extractor.conf --samp-freq=16000 --mfcc-config=/home/models/kaldi-generic-en-tdnn_sp-r20190227/conf/mfcc.conf \
/home/models/kaldi-generic-en-tdnn_sp-r20190227/model/final.mdl /home/models/kaldi-generic-en-tdnn_sp-r20190227/model/graph/HCLG.fst /home/models/kaldi-generic-en-tdnn_sp-r20190227/model/graph/words.txt

And I keep getting the following error:
ERROR (online2-tcp-nnet3-decode-faster[5.5.313~1-203c]:OnlineTransform():online-feature.cc:521) Dimension mismatch: source features have dimension 91 and LDA #cols is 280
The same arguments work fine with all the other models I use (not from Zamia). Do you know why that happens?
do you have the online.conf that you use with kaldi?

I tried to run pre trained model
./kaldi-generic-en-tdnn_fl-r20190609/'

but i am getting run time error.

M-AILABS' website m-ailabs.bayern seems offline (hostname doesn't seem to resolve). The corpus seems to be available at caito.de. Maybe should we add a link to https://www.caito.de/2019/01/the-m-ailabs-speech-dataset/ in README.md ?

can we add processing file for mozilla cv corpus version 2? Thanks

Hi Guenter,
I found some project and want to use German model from this project. But it looks like the model have some different structure (it was configured to work with Kaldi Gstreamer Server software). Сan you give any instructions or tips on how to set up this model for work with zamia-speech? I'm trying to use it in Live Mic Demo.

OS: Debian 9 Stretch 64bit
Expected behavior: Factorized ASR models work the same way as the previous models
Observed: Previous models work fine, factorized ones (both de and en) create a RuntimeError

I've been trying to use the factorized TDNN models but getting a RuntimeError in the wave decoder test.
Here is the command I've used:

python kaldi_decode_wav.py -v -m /opt/kaldi/model/kaldi-generic-de-tdnn_f test.wav 

and here is the error message:

Traceback (most recent call last): File "kaldi_decode_wav.py", line 60, in <module> kaldi_model = KaldiNNet3OnlineModel (options.modeldir, acoustic_scale=1.0, beam=7.0, frame_subsampling_factor=3) File "kaldiasr/nnet3.pyx", line 134, in kaldiasr.nnet3.KaldiNNet3OnlineModel.__cinit__ (kaldiasr/nnet3.cpp:3549) RuntimeError

What I've tried to far is to update the Debian packages (python-kaldiasr python-nltools, no updates where available) and the kaldi_decode_wav.py.

Hello,
I have just discovered zamia-speech and trying to get the live mic demo running. Everything works just fine, however, the decoding suddenly stops after few words (usually something around 10). Do you have any suggestions why that could be?

In the https://github.com/gooofy/zamia-speech#model-adaptation
you mentioned, and https://github.com/gooofy/kaldi-adapt-lm
there seemed only limited to some existing vocabulary.
Is there some configuration that can use my own vocabulary as well as LM?
thanks

Also, have a trouble understanding the sampa format, current kaldi seemed only use CMU phoneset . Is there a way that I can use the kaldi trained g2p model (trained with CMU phoneset)?

Hi,

quick question about your data sources: after pulling voxforge data and downloading TUDA audio files (voxforge forums), I still run into problems compiling the kaldi models because of missing wav files.

I checked the transcripts.csv file and it contains roughly 130000 utterances. Checking file names against wav files that are available after pulling all the data it turns out there are around 88000 missing wavs.

So my question is: Is there any chance that there's another source of utterance recordings that's not pulled automatically?

Thanks in advance.

I followed the documentation and was quite successful in replicating. But I could not find the file ./run-nnet3.sh mentioned in section https://github.com/gooofy/speech#nnet3-models .

The sequitur model does not seem able to deal with words containing umlauts (e.g. "fünfzig") when I call sequitur_gen_ipa() directly and I couldn't determine how zamia-speech deals with this.
Is there a way to autmatically generate phonemes for words with umlauts?

Hi,
Watching your discussion on "Optimizing nnet3/chain models for speed and memory consumption" really helped me a lot. Because my decoding is just limited in a small situation, I tried to built a small HCLG.fst so that the decoding will be faster. But as I followed the steps in Online decoding in Kaldi-Example for using your own language model with existing online-nnet2 models , I found the already-built model missing the "tree" file. Could you please share it which is expected in the "tdnn_250" and "tdnn_sp"?

Thanks in advance. ^_^

Hi
It seems environment.yml is too old and this demo:
http://goofy.zamia.org/zamia-speech/misc/kaldi_decode_live.py
isn't compatible with it.
Is there any plan to update this file with new version of compatible libraries?

Hi,

I would like adding a new word, but this word doesn't exist in CMU pronounce dictionary. So I used logios lexicon tool to generate a new pronunciation dictionary. I would like to know how to add this new dictionary to the existing model? Thank you so much.

hello~
i made kaldi decode live server with my pc and
i made my android application to send my voice to server with socket.
i'd like to use socket streaming record data not pulseaudio from pc.
please help me

my code is like below. it did not work
--> samples=record(sc)
--> audio, finalize = vad.process_audio(samples)

thank you!

    s.setsockopt(socket.SOL_SOCKET, socket.SO_REUSEADDR, 1)
    s.bind((HOST, PORT))
    s.listen(10)
    while(True):
        print "Listening at:", s.getsockname()
        sc, addr=s.accept()
        print "Connection established with:", addr
       

        while True:
            samples=record(sc)
            audio, finalize = vad.process_audio(samples)
	    if not audio:
		continue
	    user_utt, confidence = asr.decode(audio, finalize, stream_id=STREAM_ID)

	    print "\r%s                     " % user_utt,

	    if finalize:
		print

Do you have any docker images for this?

I have been following the Github Quickstart link which converts 4 demo wavs files to text.
It works fine, but now when I use my own Wav file it throughs an error as below:

Traceback (most recent call last): File "kaldi_decode_wav.py", line 72, in <module> if decoder.decode_wav_file(wavfile): File "kaldiasr/nnet3.pyx", line 207, in kaldiasr.nnet3.KaldiNNet3OnlineDecoder.decode_wav_file (kaldiasr/nnet3.cpp:4726) File "kaldiasr/nnet3.pyx", line 170, in kaldiasr.nnet3.KaldiNNet3OnlineDecoder.decode (kaldiasr/nnet3.cpp:3968) RuntimeError`

The file I am using is a vimeo video converted to wav using youtube-dl.
get the wav file using this command

youtube-dl --extract-audio --audio-format wav https://vimeo.com/73643788

I give this file as input to the kaldi_decode_wav.py

Can anyone help me what thing I am doing wrong?

Have you tried running on the android side?

I have been trying to run the adaptation script. I followed the instructions but the script fails at multiple stages. I tried to correct but then gave up after dozens of errors. Has anybody ran it successfully?

Hey I noticed that you don't have instructions for running this on Windows. It is possible?

Many thanks,
Kiran

I have a question regarding the performance (in terms of accuracy) of kaldi models (specifically kaldi-generic-en-tdnn_f) versus other well-known engines from Google and Watson when doing speech-to-text. From my preliminary testing (on telephony data), kaldi models are not nearly as accurate as that of Google and Watson. My typical audio file is ~ 30-60 secs long.

Since I'm not sure how this model was trained, should I chunk the audio file to multiple files at sentence length to improve the performance ? Has anyone had good luck with Kaldi in terms of performance as compared to Google and Watson?

Hi
I am trying to map the non-silenence phones to IPA or Arpabet
I tried this but it is not complete
https://github.com/gooofy/py-nltools/blob/master/nltools/phonetics.py

What does the '
represent

Thank you

I prepared the dataset similar to zamia-en dataset with 3000 wav files and prompts.
1.How to fine tune existing acoustic model
2.How to train the acoustic model from scratch

Hi,

I am trying to adapt a model but running into some trouble when invoking the "run-adaption.sh" script. It fails to call the script "utils/prepare_lang.sh" and others like "utils/mkgraph.sh" as well.
After some investigation I assume this is what is missing:
https://github.com/kaldi-asr/kaldi/tree/master/egs/wsj/s5/utils

I made a simple raspberry setup according to the instructions in the readme. I get the /opt/kaldi folder but obviously some directories like "egs" are missing there.

Is this a bug, should it work out of the box or are there others things that should be installed/configured first?

in the lexicon of the downloaded asr models, the phonems for nspc are
nC
nspc nC

how ever the nC does not exist, shouldn't this be SPN?
Does nspc represent in the kaldi standard recipes

The repo describes ways to use LibriVox. Hence, the following data release derived from LibriVox might be interesting: http://www.m-ailabs.bayern/en/the-mailabs-speech-dataset/
The audio files are segmented (by sentence) and transcribed. For German, the files amount to 237 hours, but they also provide data for the following languages: en_UK en_US es_ES it_IT uk_UK ru_RU fr_FR pl_PL

I retrained a German model, but it runs out of GPU RAM in:
nnet3-chain-train --use-gpu=wait --apply-deriv-weights=False --l2-regularize=5e-05 --leaky-hmm-coefficient=0.1 --read-cache=exp/nnet3_chain/tdnn_250/cache.1 --write-cache=exp/nnet3_chain/tdnn_250/cache.2 --xent-regularize=0.1 --print-interval=10 --momentum=0.0 --max-param-change=2.0 --backstitch-training-scale=0.0 --backstitch-training-interval=1 --l2-regularize-factor=1.0 --srand=1 "nnet3-am-copy --raw=true --learning-rate=0.000998730064726 --scale=0.980025398705 exp/nnet3_chain/tdnn_250/1.mdl - |" exp/nnet3_chain/tdnn_250/den.fst "ark,bg:nnet3-chain-copy-egs --frame-shift=2 ark:exp/nnet3_chain/tdnn_250/egs/cegs.2.ark ark:- | nnet3-chain-shuffle-egs --buffer-size=5000 --srand=1 ark:- ark:- | nnet3-chain-merge-egs --minibatch-size=512 ark:- ark:- |" exp/nnet3_chain/tdnn_250/2.1.raw

I have 4 GB GPU RAM, which is only used by kaldi in exclusive mode. This was enough back in June. What are your experiences or recommendations?

The error message contains this:
ERROR (nnet3-chain-train[5.5.95~1-4bdb]:AllocateNewRegion():cu-allocator.cc:513) Failed to allocate a memory region of 356515840 bytes. Possibly smaller minibatch size would help. Memory info: free:190M, used:3844M, total:4035M, free/total:0.0473123

So, should I retry with --minibatch-size=384 (instead of 512)? This value makes this step complete, but I probably have to rerun all steps to be sure.

Hi, I tried to make a new and clean setup on a raspberry just like I did 3 weeks ago (simply with the apt get command). But now I suddenly get this error message:

pi@raspberrypi:~ $ python kaldi_decode_wav.py -v demo?.wav
Traceback (most recent call last):
  File "kaldi_decode_wav.py", line 36, in <module>
    from kaldiasr.nnet3 import KaldiNNet3OnlineModel, KaldiNNet3OnlineDecoder
ImportError: libkaldi-decoder.so: cannot open shared object file: No such file or directory

Possibly one of those raspbian packages got broken during one of the latest commits?

I used the G2P to generate pronunciation for OOV, but the symbols its generate are not in the original dictionary

When executing speech_kaldi_export.py --add-all an exception is thrown indicating that sequitur_gen_ipa() is called with the wrong number of arguments. Indeed, it is called in speech_kaldi_export.py with 1 argument, but according to sequiturclient.py it expects 2.

Hi,

I'm computing the features for kaldi-generic-en-tdnn_f-r20190227 model.
compute-mfcc-feats
--config=kaldi-generic-en-tdnn_f-r20190227/conf/mfcc.conf
scp:transcripts/wav.scp
ark,scp:transcripts/feats.ark,transcripts/feats.scp
and running
nnet3-latgen-faster
--word-symbol-table=kaldi-generic-en-tdnn_f-r20190227/model/graph/words.txt
kaldi-generic-en-tdnn_f-r20190227/model/final.mdl
kaldi-generic-en-tdnn_f-r20190227/model/graph/HCLG.fst
ark:transcripts/feats.ark
ark,t:transcripts/lattices.ark;
Getting the following error:
ERROR (nnet3-latgen-faster[5.5.337~1-35f96]:EnsureFrameIsComputed():nnet-am-decodable-simple.cc:101) Neural net expects 'input' features with dimension 40 but you provided 13

Both files https://github.com/gooofy/speech/blob/9735ff36e47a48ae22714863bf7815b5d81552aa/data/src/speech/kaldi-run-lm.sh#L50 and https://github.com/gooofy/speech/blob/9735ff36e47a48ae22714863bf7815b5d81552aa/data/src/speech/kaldi-run-chain.sh#L69 contain the line

utils/prepare_lang.sh data/local/dict "nspc" data/local/lang data/lang

Isn't it enough if only kaldi-run-lm.sh prepares the language?

I have been trying to do force alignment with steps/nnet3/align.sh but the alignment is really off

Is there any comments about this, i did the same steps to the other models in kaldi and they are all fine

The new extraction of pronunciation information from Wiktionary delivers an impressive lexicon!

Two minor remarks:

1. Some forms are missing like "staatsflagge", "staatskonzern" although the plural forms are extracted correctly.

2. Some Wiktionary entries seem to be better than old entries in dict.ipa, e.g. "staatseigentum" and "staatseigentums" in dict.ipa currently have different IPA signs for the last "e" (and "u") while the Wiktionary versions seem to be ok.

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

I am sorry for opening another issue, but run-chain.sh complains about the missing noisy files like data/speech/de/16kHz/noisyManu-20140324-m19_deM19-32.wav and stops. Should I avoid the noisy files currently or should I try to generate them? In both cases, I will need a small hint how to proceed.
Thanks.

KenLM is apparently used to adapt kaldi models in project https://github.com/gooofy/kaldi-adapt-lm, yet SRILM is used in speech_build_lm.py. What is the rationale? Cannot KenLM be used in speech_build_lm.py phase instead?

What are the default values for w2l_env_activate,w2l_train ? they are missing from speechrc file.

./wav2letter_export.py -l en -v generic-en dict-en.ipa generic_en_lang_model_medium voxforge_en cv_corpus_v1 librispeech ljspeech m_ailabs_en tedlium3 zamia_en voxforge_en_noisy librispeech_noisy cv_corpus_v1_noisy cv_corpus_v1_phone zamia_en_noisy voxforge_en_phone librispeech_phone zamia_en_phone
Traceback (most recent call last):
File "./wav2letter_export.py", line 98, in
w2l_env_activate = config.get("speech", "w2l_env_activate")
File "/usr/lib/python2.7/ConfigParser.py", line 618, in get
raise NoOptionError(option, section)
ConfigParser.NoOptionError: No option 'w2l_env_activate' in section: 'speech'

Hi Guenter!

I was just checking out the data you link to in this section: https://github.com/gooofy/zamia-speech#speech-corpora

I downloaded the Forschergeist data, but I don't see licensing or info on where it's from... did I miss it?

Very cool work you have here:)

-josh