Author: Toktam Zoughi
DATE: 2024-7-10

Running ASR (Automatic Speech Recognition) step by step:

1. Go to the this link and follow INSTALL instructions there: https://github.com/kaldi-asr/kaldi
2. Git clone "take clone address from gitlab site(Press the clone button)"
3. Cd full_asr folder (the downloaded directory)
4. To have DataBase files run "dvc pull"
5. Cd to "for_Finglish" directory
6. Set the paths located at the "for_Finglish/path.sh" script
7. Run "NewRun_conv.sh" script (like: ./NewRun_conv.sh)
8. If you run this script successfully, it will create the following directory: "$saved_run/exp$id/nnet3/tdnn_lstm_lfr1a_sp_online" where "$saved_run" is the variable which is set on "for_Finglish/path.sh" script and "$id" is the variable which is set on "timit_s5/NewRun_Hiva.sh" script (This step may be completed in 6 days).
   

Notice: If you don't need to test the online model do not proceed the following steps.

9. Copy the "tdnn_lstm_lfr1a_sp_online" from the previous step and paste it to the "6service/exp"

10. You should have a wave file on 6service file named "mytest.wav"

11. Open two screen on your "Mobaxterm"

12. Run the following command on the first screen:

    #Now, we can compile the decoding graph with the new language model, using the following command model_dir=/mnt/HDD2/zoughi/scripts/6service/exp/tdnn_lstm_lfr1a_sp_online/

    #Now, we can compile the decoding graph with the new language model, using the following command model_dir=/mnt/HDD2/zoughi/scripts/6service/exp/tdnn_lstm_lfr1a_sp_online/ graph_own_dir=$model_dir/graph

    #online2-tcp-nnet3-decode-faster /mnt/HDD2/zoughi/scripts/6service/utils2/online2-tcp-nnet3-decode-faster --samp-freq=16000 --frames-per-chunk=20 --extra-left-context-initial=0
    --frame-subsampling-factor=3 --config=$model_dir/conf/online.conf --min-active=200 --max-active=7000
    --beam=15.0 --lattice-beam=6.0 --acoustic-scale=1.0 --port-num=5065 $model_dir/final.mdl $graph_own_dir/HCLG.fst $graph_own_dir/words.txt

13. Run the following commands on the second screen:

    #/mnt/HDD2/zoughi/scripts/6service/utils2/sox mytest.wav -t raw -c 1 -b 16 -r 16k -e signed-integer - | nc -N localhost 5065 /mnt/HDD2/zoughi/scripts/6service/utils2/sox mytest.wav -t raw -c 1 -b 16 -r 16k -e signed-integer - | nc -N localhost 5065 > /mnt/HDD2/zoughi/scripts/6service/data/out1.txt

    #********************** THIS is python code **************************** python3.7 import sys, subprocess subprocess.call(['/usr/bin/perl', '/mnt/HDD2/zoughi/scripts/6service/utils2/En2Per2.pl'])

14. If you run all of this steps successfully, you should see the corresponding text of "mytest.wav" on the second screen.

By: Toktam Zoughi