DeepSpeech Notes

lm.binary was generated from the LibriSpeech normalized LM training text, available {http://www.openslr.org/11}}, following this recipe (Jupyter notebook code):


import gzip
import io
import os

from urllib import request

# Grab corpus.
url = 'http://www.openslr.org/resources/11/librispeech-lm-norm.txt.gz'
data_upper = '/tmp/upper.txt.gz'
request.urlretrieve(url, data_upper)

# Convert to lowercase and cleanup.
data_lower = '/tmp/lower.txt'
with open(data_lower, 'w', encoding='utf-8') as lower:
    with io.TextIOWrapper(io.BufferedReader(gzip.open(data_upper)), encoding='utf8') as upper:
        for line in upper:
            lower.write(line.lower())

# Build pruned LM.
lm_path = '/tmp/lm.arpa'
!lmplz --order 5 \
       --temp_prefix /tmp/ \
       --memory 50% \
       --text {data_lower} \
       --arpa {lm_path} \
       --prune 0 0 0 1

# Quantize and produce trie binary.
binary_path = '/tmp/lm.binary'
!build_binary -a 255 \
              -q 8 \
              trie \
              {lm_path} \
              {binary_path} 
os.remove(lm_path)

The trie was then generated from the vocabulary of the language model:

./generate_trie ../data/alphabet.txt /tmp/lm.binary /tmp/trie



-------------------------------------------------

Generating TRIE and LM files in DeepSpeech:


/DeepSpeech/native_client/kenlm/build/bin#  ./lmplz --text /docker_files/lm_own_v3/lower.txt --arpa /docker_files/lm_own_v3/words.arpa --o 5 --prune 0 0 0 1


/DeepSpeech/native_client# ./generate_trie /docker_files/lm_own_v3/alphabet.txt /docker_files/lm_own_v3/lm_own_v3.binary /docker_files/lm_own_v3/trie

/DeepSpeech/native_client/kenlm/build/bin#  ./build_binary -s -a 255 -q 8 trie /docker_files/lm_own_v3/words.arpa /docker_files/lm_own_v3/lm_own_v3.binary



./deepspeech --model /docker_files/checkpoints_cv_mozilla/output_graph.pb --alphabet /docker_files/lm_own_v3/alphabet.txt --lm /docker_files/lm_own_v3/lm_own_v3.binary  --trie /docker_files/lm_own_v3/trie --audio /docker_files/wavs/wavs/preddy8_india.wav




------------------------------------------------------------
Training commnand for tranfer learning

python DeepSpeech.py --train_files /docker_files/CV_own_2/custom_train.csv --test_files /docker_files/CV_own_2/custom_test.csv --dev_files /docker_files/CV_own_2/custom_dev.csv --drop_source_layers 2 --source_model_checkpoint_dir /docker_files/checkpoints_cv_mozilla/ --checkpoint_dir /docker_files/checkpoints_cv_mozilla/ --export_dir /docker_files/checkpoints_cv_mozilla/ --alphabet_config_path /docker_files/lm_own_v3/alphabet.txt --lm_binary_path /docker_files/lm_own_v3/lm_own_v3.binary --lm_trie_path /docker_files/lm_own_v3/trie --epoch 3


python DeepSpeech.py --train_files /docker_files/CV_own_2/custom_train_CVSVA.csv --test_files /docker_files/CV_own_2/custom_test_CVSVA.csv --dev_files /docker_files/CV_own_2/custom_dev_CVSVA.csv --drop_source_layers 3 --source_model_checkpoint_dir /docker_files/checkpoints_cv_mozilla/ --checkpoint_dir /docker_files/checkpoints_cv_mozilla/ --export_dir /docker_files/checkpoints_cv_mozilla/ --alphabet_config_path /docker_files/lm_own_v3/alphabet.txt --lm_binary_path /docker_files/lm_own_v3/lm_own_v3.binary --lm_trie_path /docker_files/lm_own_v3/trie --epoch -10 --train_batch_size 8


--------------------------------------------------------------
Starting the docker:
#Inside mycroft-core folder

docker run -it --rm --device /dev/snd -e PULSE_SERVER=unix:${XDG_RUNTIME_DIR}/pulse/native -v ${XDG_RUNTIME_DIR}/pulse/native:${XDG_RUNTIME_DIR}/pulse/native -v ~/.config/pulse/cookie:/root/.config/pulse/cookie -v /home/admin1/0Pratesh/Speech/mycroft-core:/root/ -v /home/admin1/0Pratesh/Speech/deepspeech/frozen_models/:/models/  mycroft_core:ubuntu16_mic bash

#Inside the docker:
apt-get update
    7  apt-get install sudo
    8  useradd -m docker && echo "docker:docker" | chpasswd && adduser docker sudo

apt-get install -y git python3 python3-dev python-setuptools python-gobject-2-dev libtool libffi-dev libssl-dev autoconf automake bison swig libglib2.0-dev portaudio19-dev mpg123 screen flac curl libicu-dev pkg-config automake libjpeg-dev libfann-dev build-essential jq
   20  apt-get install -y git 
apt-get install vim
   47  vim /etc/pip.conf


apt-get install pulseaudio alsa-utils -y
git ssl.Verify=false
git config --global http.sslverify false


#Activate the venv
cd /root/
source .venv/bin/activate

git clone https://git.visteon.com/SmartVoiceAssistant/mycroft-skills.git

Bringing up mycroft:
vim mycroft/configuration/mycroft.conf
--> search for deepspeech you'll understand where to change for frozen,lm and trie file

#To start mycroft
./dev_setup.sh --allow-root (not needed everytime you start docker)
./start-mycroft.sh debug



#To pull the latest mycrodt branch.. set the gitlab credentials
git config --global user.email "preddy8@visteon.com"
git config --global user.name "pratesh"


git stash
git pull
git checkout development
git pusll
git pull
git checkout 9-download-skills-with-tags-or-branch-2


---------------------------------------------------------------------------
Disowning a docker container:

Run the docker with -dt flag 
-d for daemon mode

docker run -dt $IMAGE_NAME

This will run it as daemon.Now exec this docker container

docker exec -it $CONT_ID bash 

Inside docker run what ever process u want but by disowning it 

nohup $$$$$PROCESS$$$$ & 
#put & at the end to make it a background process
#Ad nohup in front to disown the process

Now exit the container.

DOcker will still running in the remote machine.

You can enter this container any time (direclty or through SSH)