Comments (1)
How to train the model?
I have run the run.sh for two days, but it seems there are some mistakes from the log. This log stayed more than 31 hours. I am afraid that the trianning is ended. The log is as follows.
steps/decode.sh --nj 20 --cmd run.pl --max-jobs-run 10 exp/tri1/graph_nosp_tgsmall data/dev_clean exp/tri1/decode_nosp_tgsmall_dev_clean
decode.sh: feature type is delta
steps/lmrescore_const_arpa.sh --cmd run.pl --max-jobs-run 10 data/lang_nosp_test_tgsmall data/lang_nosp_test_tglarge data/test_other exp/tri4b/decode_nosp_tgsmall_test_other exp/tri4b/decode_nosp_tglarge_test_other
steps/lmrescore_const_arpa.sh --cmd run.pl --max-jobs-run 10 data/lang_nosp_test_tgsmall data/lang_nosp_test_fglarge data/test_other exp/tri4b/decode_nosp_tgsmall_test_other exp/tri4b/decode_nosp_fglarge_test_other
utils/build_const_arpa_lm.sh: line 47: 28623 Killed arpa-to-const-arpa --bos-symbol=$bos --eos-symbol=$eos --unk-symbol=$unk "gunzip -c $arpa_lm | utils/map_arpa_lm.pl $new_lang/words.txt|" $new_lang/G.carpa
root@yue-W580-G20:/home/yue/Desktop/yxj/kaldi-master/egs/librispeech/news5# run.pl: 10 / 20 failed, log is in exp/tri4b/decode_nosp_fglarge_test_other/log/rescorelm.*.log
steps/diagnostic/analyze_lats.sh --cmd run.pl --max-jobs-run 10 exp/tri3b/graph_nosp_tgsmall exp/tri3b/decode_nosp_tgsmall_dev_other.si
steps/diagnostic/analyze_lats.sh: see stats in exp/tri3b/decode_nosp_tgsmall_dev_other.si/log/analyze_alignments.log
Overall, lattice depth (10,50,90-percentile)=(2,15,69) and mean=28.1
steps/diagnostic/analyze_lats.sh: see stats in exp/tri3b/decode_nosp_tgsmall_dev_other.si/log/analyze_lattice_depth_stats.log
steps/decode_fmllr.sh: feature type is lda
steps/decode_fmllr.sh: getting first-pass fMLLR transforms.
steps/decode_fmllr.sh: doing main lattice generation phase
steps/diagnostic/analyze_lats.sh --cmd run.pl --max-jobs-run 10 exp/tri1/graph_nosp_tgsmall exp/tri1/decode_nosp_tgsmall_dev_clean
steps/diagnostic/analyze_lats.sh: see stats in exp/tri1/decode_nosp_tgsmall_dev_clean/log/analyze_alignments.log
Overall, lattice depth (10,50,90-percentile)=(1,6,43) and mean=17.3
steps/diagnostic/analyze_lats.sh: see stats in exp/tri1/decode_nosp_tgsmall_dev_clean/log/analyze_lattice_depth_stats.log
steps/lmrescore.sh --cmd run.pl --max-jobs-run 10 data/lang_nosp_test_tgsmall data/lang_nosp_test_tgmed data/dev_clean exp/tri1/decode_nosp_tgsmall_dev_clean exp/tri1/decode_nosp_tgmed_dev_clean
fstdeterminizestar
fstrmsymbols data/lang_nosp_test_tgmed/phones/disambig.int
steps/lmrescore_const_arpa.sh --cmd run.pl --max-jobs-run 10 data/lang_nosp_test_tgsmall data/lang_nosp_test_tglarge data/dev_clean exp/tri1/decode_nosp_tgsmall_dev_clean exp/tri1/decode_nosp_tglarge_dev_clean
steps/decode.sh --nj 20 --cmd run.pl --max-jobs-run 10 exp/tri1/graph_nosp_tgsmall data/dev_other exp/tri1/decode_nosp_tgsmall_dev_other
decode.sh: feature type is delta
steps/decode_fmllr.sh: estimating fMLLR transforms a second time.
steps/decode_fmllr.sh: doing a final pass of acoustic rescoring.
steps/diagnostic/analyze_lats.sh --cmd run.pl --max-jobs-run 10 exp/tri3b/graph_nosp_tgsmall exp/tri3b/decode_nosp_tgsmall_dev_other
steps/diagnostic/analyze_lats.sh: see stats in exp/tri3b/decode_nosp_tgsmall_dev_other/log/analyze_alignments.log
Overall, lattice depth (10,50,90-percentile)=(1,8,48) and mean=19.2
steps/diagnostic/analyze_lats.sh: see stats in exp/tri3b/decode_nosp_tgsmall_dev_other/log/analyze_lattice_depth_stats.log
steps/lmrescore.sh --cmd run.pl --max-jobs-run 10 data/lang_nosp_test_tgsmall data/lang_nosp_test_tgmed data/dev_other exp/tri3b/decode_nosp_tgsmall_dev_other exp/tri3b/decode_nosp_tgmed_dev_other
fstdeterminizestar
fstrmsymbols data/lang_nosp_test_tgmed/phones/disambig.int
steps/diagnostic/analyze_lats.sh --cmd run.pl --max-jobs-run 10 exp/tri2b/graph_nosp_tgsmall exp/tri2b/decode_nosp_tgsmall_dev_other
steps/diagnostic/analyze_lats.sh: see stats in exp/tri2b/decode_nosp_tgsmall_dev_other/log/analyze_alignments.log
Overall, lattice depth (10,50,90-percentile)=(2,19,111) and mean=43.3
steps/diagnostic/analyze_lats.sh: see stats in exp/tri2b/decode_nosp_tgsmall_dev_other/log/analyze_lattice_depth_stats.log
steps/lmrescore.sh --cmd run.pl --max-jobs-run 10 data/lang_nosp_test_tgsmall data/lang_nosp_test_tgmed data/dev_other exp/tri2b/decode_nosp_tgsmall_dev_other exp/tri2b/decode_nosp_tgmed_dev_other
fstdeterminizestar
fstrmsymbols data/lang_nosp_test_tgmed/phones/disambig.int
steps/lmrescore_const_arpa.sh --cmd run.pl --max-jobs-run 10 data/lang_nosp_test_tgsmall data/lang_nosp_test_tglarge data/dev_other exp/tri3b/decode_nosp_tgsmall_dev_other exp/tri3b/decode_nosp_tglarge_dev_other
steps/diagnostic/analyze_lats.sh --cmd run.pl --max-jobs-run 10 exp/mono/graph_nosp_tgsmall exp/mono/decode_nosp_tgsmall_dev_clean
steps/diagnostic/analyze_lats.sh: see stats in exp/mono/decode_nosp_tgsmall_dev_clean/log/analyze_alignments.log
Overall, lattice depth (10,50,90-percentile)=(5,53,299) and mean=114.4
steps/diagnostic/analyze_lats.sh: see stats in exp/mono/decode_nosp_tgsmall_dev_clean/log/analyze_lattice_depth_stats.log
steps/decode.sh --nj 20 --cmd run.pl --max-jobs-run 10 exp/mono/graph_nosp_tgsmall data/dev_other exp/mono/decode_nosp_tgsmall_dev_other
decode.sh: feature type is delta
steps/lmrescore_const_arpa.sh --cmd run.pl --max-jobs-run 10 data/lang_nosp_test_tgsmall data/lang_nosp_test_tglarge data/dev_other exp/tri2b/decode_nosp_tgsmall_dev_other exp/tri2b/decode_nosp_tglarge_dev_other
steps/diagnostic/analyze_lats.sh --cmd run.pl --max-jobs-run 10 exp/tri1/graph_nosp_tgsmall exp/tri1/decode_nosp_tgsmall_dev_other
steps/diagnostic/analyze_lats.sh: see stats in exp/tri1/decode_nosp_tgsmall_dev_other/log/analyze_alignments.log
Overall, lattice depth (10,50,90-percentile)=(3,29,155) and mean=60.8
steps/diagnostic/analyze_lats.sh: see stats in exp/tri1/decode_nosp_tgsmall_dev_other/log/analyze_lattice_depth_stats.log
steps/lmrescore.sh --cmd run.pl --max-jobs-run 10 data/lang_nosp_test_tgsmall data/lang_nosp_test_tgmed data/dev_other exp/tri1/decode_nosp_tgsmall_dev_other exp/tri1/decode_nosp_tgmed_dev_other
fstrmsymbols data/lang_nosp_test_tgmed/phones/disambig.int
fstdeterminizestar
steps/lmrescore_const_arpa.sh --cmd run.pl --max-jobs-run 10 data/lang_nosp_test_tgsmall data/lang_nosp_test_tglarge data/dev_other exp/tri1/decode_nosp_tgsmall_dev_other exp/tri1/decode_nosp_tglarge_dev_other
steps/diagnostic/analyze_lats.sh --cmd run.pl --max-jobs-run 10 exp/mono/graph_nosp_tgsmall exp/mono/decode_nosp_tgsmall_dev_other
steps/diagnostic/analyze_lats.sh: see stats in exp/mono/decode_nosp_tgsmall_dev_other/log/analyze_alignments.log
Overall, lattice depth (10,50,90-percentile)=(15,124,500) and mean=206.1
steps/diagnostic/analyze_lats.sh: see stats in exp/mono/decode_nosp_tgsmall_dev_other/log/analyze_lattice_depth_stats.log
from alibaba-mit-speech.
Related Issues (20)
- run_fsmn_ivector.sh Value of $cuda_cmd
- what if i have no gpu, how long it will take to train this model in kaldi HOT 5
- where to download the trained dfsmn model
- Error: in data_fbank/train_960_cleaned, recording-ids extracted from wav.scp and reco2dur file differ HOT 1
- I got a stun when I am excuting "local/nnet/run_fsmn_ivector.sh DFSMN_S" HOT 1
- when dfsmn support muti GPU HOT 1
- What is the cause of this error? HOT 2
- low GPU memory usage HOT 1
- which nnet version DFSMN uses
- 有没有准确率较高的中文模型提供试用下啊 HOT 1
- DFSMN does not support CUDA 10
- 运行local/nnet/run_fsmn.sh DFSMN_L中的CE-training时出错
- git am --signoff < /data/glusterfs_speech_04/11085090/Alibaba-MIT-Speech/Alibaba_MIT_Speech_DFSMN.patch
- Run_fsmn_invector.sh have some errors,please give me right script file,thanks
- 阿里大帝,中文模型有没有? HOT 1
- No file nnet-train-fsmn-streams in src/nnetbin?
- where is fbank.cfg? HOT 3
- nnet-train-fsmn-streams: command not found HOT 1
- run.sh error HOT 1
Recommend Projects
-
React
A declarative, efficient, and flexible JavaScript library for building user interfaces.
-
Vue.js
🖖 Vue.js is a progressive, incrementally-adoptable JavaScript framework for building UI on the web.
-
Typescript
TypeScript is a superset of JavaScript that compiles to clean JavaScript output.
-
TensorFlow
An Open Source Machine Learning Framework for Everyone
-
Django
The Web framework for perfectionists with deadlines.
-
Laravel
A PHP framework for web artisans
-
D3
Bring data to life with SVG, Canvas and HTML. 📊📈🎉
-
Recommend Topics
-
javascript
JavaScript (JS) is a lightweight interpreted programming language with first-class functions.
-
web
Some thing interesting about web. New door for the world.
-
server
A server is a program made to process requests and deliver data to clients.
-
Machine learning
Machine learning is a way of modeling and interpreting data that allows a piece of software to respond intelligently.
-
Visualization
Some thing interesting about visualization, use data art
-
Game
Some thing interesting about game, make everyone happy.
Recommend Org
-
Facebook
We are working to build community through open source technology. NB: members must have two-factor auth.
-
Microsoft
Open source projects and samples from Microsoft.
-
Google
Google ❤️ Open Source for everyone.
-
Alibaba
Alibaba Open Source for everyone
-
D3
Data-Driven Documents codes.
-
Tencent
China tencent open source team.
from alibaba-mit-speech.