[speechx] rm ds2 && rm boost (#2786)

* fix openfst download error * add acknowledgments of openfst * refactor directory * clean ctc_decoders dir * add nnet cache && make 2 thread work * do not compile websocket * rm ds2 && rm boost * rm ds2 example

[speechx] rm ds2 && rm boost (#2786)
* fix openfst download error * add acknowledgments of openfst * refactor directory * clean ctc_decoders dir * add nnet cache && make 2 thread work * do not compile websocket * rm ds2 && rm boost * rm ds2 example
acf1d272 · YangZhou · GitHub · 5046d8ee · acf1d272 · acf1d272
66 changed file
--- a/.pre-commit-config.yaml
+++ b/.pre-commit-config.yaml
@@ -57,13 +57,13 @@ repos:
        entry: bash .pre-commit-hooks/clang-format.hook -i
        language: system
        files: \.(h\+\+|h|hh|hxx|hpp|cuh|c|cc|cpp|cu|c\+\+|cxx|tpp|txx)$
-        exclude: (?=speechx/speechx/kaldi|audio/paddleaudio/src|speechx/patch|speechx/tools/fstbin|speechx/tools/lmbin|third_party/ctc_decoders).*(\.cpp|\.cc|\.h|\.hpp|\.py)$ 
+        exclude: (?=speechx/speechx/kaldi|audio/paddleaudio/src|speechx/patch|speechx/tools/fstbin|speechx/tools/lmbin|third_party/ctc_decoders|speechx/speechx/common/utils).*(\.cpp|\.cc|\.h|\.hpp|\.py)$ 
    -   id: cpplint
        name: cpplint
        description: Static code analysis of C/C++ files
        language: python
        files: \.(h\+\+|h|hh|hxx|hpp|cuh|c|cc|cpp|cu|c\+\+|cxx|tpp|txx)$
-        exclude: (?=speechx/speechx/kaldi|audio/paddleaudio/src|speechx/patch|speechx/tools/fstbin|speechx/tools/lmbin|third_party/ctc_decoders).*(\.cpp|\.cc|\.h|\.hpp|\.py)$ 
+        exclude: (?=speechx/speechx/kaldi|audio/paddleaudio/src|speechx/patch|speechx/tools/fstbin|speechx/tools/lmbin|third_party/ctc_decoders|speechx/speechx/common/utils).*(\.cpp|\.cc|\.h|\.hpp|\.py)$ 
        entry: cpplint --filter=-build,-whitespace,+whitespace/comma,-whitespace/indent
 -   repo: https://github.com/asottile/reorder_python_imports
    rev: v2.4.0

--- a/speechx/CMakeLists.txt
+++ b/speechx/CMakeLists.txt
@@ -44,9 +44,6 @@ option(TEST_DEBUG "option for debug" OFF)
 option(USE_PROFILING "enable c++ profling" OFF)
 option(WITH_TESTING "unit test" ON)
-option(USING_U2  "compile u2 model." ON)
-option(USING_DS2 "compile with ds2 model." OFF)
 option(USING_GPU "u2 compute on GPU." OFF)
 ###############################################################################
@@ -56,21 +53,6 @@ include(gflags)
 include(glog)
-# boost
-# include(boost) # not work
-set(boost_SOURCE_DIR ${fc_patch}/boost-src)
-set(BOOST_ROOT ${boost_SOURCE_DIR})
-include_directories(${boost_SOURCE_DIR})
-link_directories(${boost_SOURCE_DIR}/stage/lib)
-# Eigen
-include(eigen)
-find_package(Eigen3 REQUIRED)
-# Kenlm
-include(kenlm)
-add_dependencies(kenlm eigen boost)
 #openblas
 include(openblas)

--- a/speechx/build.sh
+++ b/speechx/build.sh
@@ -4,20 +4,5 @@ set -xe
 # the build script had verified in the paddlepaddle docker image.
 # please follow the instruction below to install PaddlePaddle image.
 # https://www.paddlepaddle.org.cn/documentation/docs/zh/install/docker/linux-docker.html 
-boost_SOURCE_DIR=$PWD/fc_patch/boost-src
+cmake -B build
-if [ ! -d ${boost_SOURCE_DIR} ]; then wget -c https://boostorg.jfrog.io/artifactory/main/release/1.75.0/source/boost_1_75_0.tar.gz 
-  tar xzfv boost_1_75_0.tar.gz
-  mkdir -p $PWD/fc_patch
-  mv boost_1_75_0 ${boost_SOURCE_DIR} 
-  cd ${boost_SOURCE_DIR}
-  bash ./bootstrap.sh
-  ./b2
-  cd -
-  echo -e "\n"
-fi
-#rm -rf build
-mkdir -p build
-cmake -B build -DBOOST_ROOT:STRING=${boost_SOURCE_DIR}
 cmake --build build -j
--- a/speechx/examples/ds2_ol/README.md
+++ b/speechx/examples/ds2_ol/README.md
-# Deepspeech2 Streaming ASR
-## Examples
-* `websocket` - Streaming ASR with websocket for deepspeech2_aishell.    
-* `aishell` - Streaming Decoding under aishell dataset, for local WER test.    
-* `onnx` - Example to convert deepspeech2 to onnx format.
--- a/speechx/examples/ds2_ol/aishell/.gitignore
+++ b/speechx/examples/ds2_ol/aishell/.gitignore
-data
-exp
-aishell_*
--- a/speechx/examples/ds2_ol/aishell/README.md
+++ b/speechx/examples/ds2_ol/aishell/README.md
-# Aishell - Deepspeech2 Streaming
-> We recommend using U2/U2++ model instead of DS2, please see [here](../../u2pp_ol/wenetspeech/).
-A C++ deployment example for using the deepspeech2 model to recognize `wav` and compute `CER`. We using AISHELL-1 as test data.
-## Source path.sh
-```bash
-. path.sh
-```
-SpeechX bins is under `echo $SPEECHX_BUILD`, more info please see `path.sh`.
-## Recognize with linear feature
-```bash
-bash run.sh
-```
-`run.sh` has multi stage, for details please see `run.sh`: 
-1. donwload dataset, model and lm
-2. convert cmvn format and compute feature
-3. decode w/o lm by feature
-4. decode w/ ngram lm by feature
-5. decode w/ TLG graph by feature
-6. recognize w/ TLG graph by wav input
-### Recognize with `.scp` file for wav
-This sciprt using `recognizer_main` to recognize wav file.
-The input is `scp` file which look like this:
-```text
-# head data/split1/1/aishell_test.scp 
-BAC009S0764W0121        /workspace/PaddleSpeech/speechx/examples/u2pp_ol/wenetspeech/data/test/S0764/BAC009S0764W0121.wav
-BAC009S0764W0122        /workspace/PaddleSpeech/speechx/examples/u2pp_ol/wenetspeech/data/test/S0764/BAC009S0764W0122.wav
-...
-BAC009S0764W0125        /workspace/PaddleSpeech/speechx/examples/u2pp_ol/wenetspeech/data/test/S0764/BAC009S0764W0125.wav
-```
-If you want to recognize one wav, you can make `scp` file like this:
-```text
-key  path/to/wav/file
-```
-Then specify `--wav_rspecifier=` param for `recognizer_main` bin. For other flags meaning, please see `help`:
-```bash
-recognizer_main --help
-```
-For the exmaple to using `recognizer_main` please see `run.sh`.
-### CTC Prefix Beam Search w/o LM
-```
-Overall -> 16.14 % N=104612 C=88190 S=16110 D=312 I=465
-Mandarin -> 16.14 % N=104612 C=88190 S=16110 D=312 I=465
-Other -> 0.00 % N=0 C=0 S=0 D=0 I=0
-```
-### CTC Prefix Beam Search w/ LM
-LM: zh_giga.no_cna_cmn.prune01244.klm
-```
-Overall -> 7.86 % N=104768 C=96865 S=7573 D=330 I=327
-Mandarin -> 7.86 % N=104768 C=96865 S=7573 D=330 I=327
-Other -> 0.00 % N=0 C=0 S=0 D=0 I=0
-```
-### CTC TLG WFST
-LM: [aishell train](http://paddlespeech.bj.bcebos.com/speechx/examples/ds2_ol/aishell/aishell_graph.zip)
--acoustic_scale=1.2
-```
-Overall -> 11.14 % N=103017 C=93363 S=9583 D=71 I=1819
-Mandarin -> 11.14 % N=103017 C=93363 S=9583 D=71 I=1818
-Other -> 0.00 % N=0 C=0 S=0 D=0 I=1
-```
-LM: [wenetspeech](http://paddlespeech.bj.bcebos.com/speechx/examples/ds2_ol/aishell/wenetspeech_graph.zip)
--acoustic_scale=1.5
-```
-Overall -> 10.93 % N=104765 C=93410 S=9780 D=1575 I=95
-Mandarin -> 10.93 % N=104762 C=93410 S=9779 D=1573 I=95
-Other -> 100.00 % N=3 C=0 S=1 D=2 I=0
-```
-## Recognize with fbank feature
-This script is same to `run.sh`, but using fbank feature.
-```bash
-bash run_fbank.sh
-```
-### CTC Prefix Beam Search w/o LM
-```
-Overall -> 10.44 % N=104765 C=94194 S=10174 D=397 I=369
-Mandarin -> 10.44 % N=104762 C=94194 S=10171 D=397 I=369
-Other -> 100.00 % N=3 C=0 S=3 D=0 I=0
-```
-### CTC Prefix Beam Search w/ LM
-LM: zh_giga.no_cna_cmn.prune01244.klm
-```
-Overall -> 5.82 % N=104765 C=99386 S=4944 D=435 I=720
-Mandarin -> 5.82 % N=104762 C=99386 S=4941 D=435 I=720
-English -> 0.00 % N=0 C=0 S=0 D=0 I=0
-```
-### CTC TLG WFST
-LM: [aishell train](https://paddlespeech.bj.bcebos.com/s2t/paddle_asr_online/aishell_graph2.zip)
-```
-Overall -> 9.58 % N=104765 C=94817 S=4326 D=5622 I=84
-Mandarin -> 9.57 % N=104762 C=94817 S=4325 D=5620 I=84
-Other -> 100.00 % N=3 C=0 S=1 D=2 I=0
-```
-## Build TLG WFST graph 
-The script is for building TLG wfst graph, depending on `srilm`, please make sure it is installed.
-For more information please see the script below.
-```bash
- bash ./local/run_build_tlg.sh
-```
--- a/speechx/examples/ds2_ol/aishell/local/aishell_train_lms.sh
+++ b/speechx/examples/ds2_ol/aishell/local/aishell_train_lms.sh
-#!/bin/bash
-# To be run from one directory above this script.
-. ./path.sh
-nj=40
-text=data/local/lm/text
-lexicon=data/local/dict/lexicon.txt
-for f in "$text" "$lexicon"; do
-  [ ! -f $x ] && echo "$0: No such file $f" && exit 1;
-done
-# Check SRILM tools
-if ! which ngram-count > /dev/null; then
-    echo "srilm tools are not found, please download it and install it from: "
-    echo "http://www.speech.sri.com/projects/srilm/download.html"
-    echo "Then add the tools to your PATH"
-    exit 1
-fi
-# This script takes no arguments.  It assumes you have already run
-# aishell_data_prep.sh.
-# It takes as input the files
-# data/local/lm/text
-# data/local/dict/lexicon.txt
-dir=data/local/lm
-mkdir -p $dir
-cleantext=$dir/text.no_oov
-# oov to <SPOKEN_NOISE>
-# lexicon line: word char0 ... charn
-# text line: utt word0 ... wordn -> line: <SPOKEN_NOISE> word0 ... wordn
-text_dir=$(dirname $text)
-split_name=$(basename $text)
-./local/split_data.sh $text_dir $text $split_name $nj
-utils/run.pl JOB=1:$nj $text_dir/split${nj}/JOB/${split_name}.no_oov.log \
-  cat ${text_dir}/split${nj}/JOB/${split_name} \| awk -v lex=$lexicon 'BEGIN{while((getline<lex) >0){ seen[$1]=1; } }
-    {for(n=1; n<=NF;n++) {  if (seen[$n]) { printf("%s ", $n); } else {printf("<SPOKEN_NOISE> ");} } printf("\n");}' \
-    \> ${text_dir}/split${nj}/JOB/${split_name}.no_oov || exit 1;
-cat ${text_dir}/split${nj}/*/${split_name}.no_oov  > $cleantext
-# compute word counts, sort in descending order
-# line: count word
-cat $cleantext | awk '{for(n=2;n<=NF;n++) print $n; }' | sort --parallel=`nproc` | uniq -c | \
-   sort --parallel=`nproc` -nr > $dir/word.counts || exit 1;
-# Get counts from acoustic training transcripts, and add  one-count
-# for each word in the lexicon (but not silence, we don't want it
-# in the LM-- we'll add it optionally later).
-cat $cleantext | awk '{for(n=2;n<=NF;n++) print $n; }' | \
-  cat - <(grep -w -v '!SIL' $lexicon | awk '{print $1}') | \
-   sort --parallel=`nproc` | uniq -c | sort --parallel=`nproc` -nr > $dir/unigram.counts || exit 1;
-# word with <s> </s>
-cat $dir/unigram.counts | awk '{print $2}' | cat - <(echo "<s>"; echo "</s>" ) > $dir/wordlist
-# hold out to compute ppl
-heldout_sent=10000 # Don't change this if you want result to be comparable with kaldi_lm results
-mkdir -p $dir
-cat $cleantext | awk '{for(n=2;n<=NF;n++){ printf $n; if(n<NF) printf " "; else print ""; }}' | \
-  head -$heldout_sent > $dir/heldout
-cat $cleantext | awk '{for(n=2;n<=NF;n++){ printf $n; if(n<NF) printf " "; else print ""; }}' | \
-  tail -n +$heldout_sent > $dir/train
-ngram-count -text $dir/train -order 3 -limit-vocab -vocab $dir/wordlist -unk \
-  -map-unk "<UNK>" -kndiscount -interpolate -lm $dir/lm.arpa
-ngram -lm $dir/lm.arpa -ppl $dir/heldout
\ No newline at end of file
--- a/speechx/examples/ds2_ol/aishell/local/run_build_tlg.sh
+++ b/speechx/examples/ds2_ol/aishell/local/run_build_tlg.sh
-#!/bin/bash
-set -eo pipefail
-. path.sh
-# attention, please replace the vocab is only for this script. 
-# different acustic model has different vocab
-ckpt_dir=data/fbank_model
-unit=$ckpt_dir/data/lang_char/vocab.txt       # vocab file, line: char/spm_pice
-model_dir=$ckpt_dir/exp/deepspeech2_online/checkpoints/
-stage=-1
-stop_stage=100
-corpus=aishell
-lexicon=data/lexicon.txt  # line: word ph0 ... phn, aishell/resource_aishell/lexicon.txt
-text=data/text            # line: utt text, aishell/data_aishell/transcript/aishell_transcript_v0.8.txt
-. utils/parse_options.sh
-data=$PWD/data
-mkdir -p $data
-if [ $stage -le -1 ] && [ $stop_stage -ge -1 ]; then
-    if [ ! -f $data/speech.ngram.zh.tar.gz ];then
-        # download ngram
-        pushd $data
-        wget -c http://paddlespeech.bj.bcebos.com/speechx/examples/ngram/zh/speech.ngram.zh.tar.gz
-        tar xvzf speech.ngram.zh.tar.gz
-        popd
-    fi
-    if [ ! -f $ckpt_dir/data/mean_std.json ]; then
-        # download model
-        mkdir -p $ckpt_dir
-        pushd $ckpt_dir
-        wget -c https://paddlespeech.bj.bcebos.com/s2t/wenetspeech/asr0/WIP1_asr0_deepspeech2_online_wenetspeech_ckpt_1.0.0a.model.tar.gz
-        tar xzfv WIP1_asr0_deepspeech2_online_wenetspeech_ckpt_1.0.0a.model.tar.gz
-        popd
-    fi
-fi
-if [ ! -f $unit ]; then
-    echo "$0: No such file $unit"
-    exit 1;
-fi
-if ! which ngram-count; then
-    # need srilm install
-    pushd $MAIN_ROOT/tools
-    make srilm.done
-    popd
-fi
-mkdir -p data/local/dict
-if [ ${stage} -le 0 ] && [ ${stop_stage} -ge 0 ]; then
-    # Prepare dict
-    # line: char/spm_pices
-    cp $unit data/local/dict/units.txt
-    if [ ! -f $lexicon ];then
-       utils/text_to_lexicon.py --has_key true --text $text --lexicon $lexicon
-        echo "Generate $lexicon from $text"
-    fi
-    # filter by vocab
-    # line: word ph0 ... phn -> line: word char0 ... charn
-    utils/fst/prepare_dict.py \
-        --unit_file $unit \
-        --in_lexicon ${lexicon} \
-        --out_lexicon data/local/dict/lexicon.txt
-fi
-lm=data/local/lm
-mkdir -p $lm
-if [ ${stage} -le 1 ] && [ ${stop_stage} -ge 1 ]; then
-    # Train ngram lm
-    cp $text $lm/text
-    local/aishell_train_lms.sh
-    echo "build LM done."
-fi
-# build TLG
-if [ ${stage} -le 2 ] && [ ${stop_stage} -ge 2 ]; then
-  # build T & L
-  utils/fst/compile_lexicon_token_fst.sh \
-      data/local/dict data/local/tmp data/local/lang
-  # build G & TLG
-  utils/fst/make_tlg.sh data/local/lm data/local/lang data/lang_test || exit 1;
-fi
-aishell_wav_scp=aishell_test.scp
-nj=40
-cmvn=$data/cmvn_fbank.ark
-wfst=$data/lang_test
-if [ ${stage} -le 3 ] && [ ${stop_stage} -ge 3 ]; then
-    if [ ! -d $data/test ]; then
-        # download test dataset
-        pushd $data
-        wget -c https://paddlespeech.bj.bcebos.com/s2t/paddle_asr_online/aishell_test.zip
-        unzip  aishell_test.zip
-        popd
-        realpath $data/test/*/*.wav > $data/wavlist
-        awk -F '/' '{ print $(NF) }' $data/wavlist | awk -F '.' '{ print $1 }' > $data/utt_id
-        paste $data/utt_id $data/wavlist > $data/$aishell_wav_scp
-    fi
-    ./local/split_data.sh $data $data/$aishell_wav_scp $aishell_wav_scp $nj
-    # convert cmvn format
-    cmvn-json2kaldi --json_file=$ckpt_dir/data/mean_std.json --cmvn_write_path=$cmvn
-fi
-wer=aishell_wer
-label_file=aishell_result
-export GLOG_logtostderr=1
-if [ ${stage} -le 4 ] && [ ${stop_stage} -ge 4 ]; then
-    #  recognize w/ TLG graph
-    utils/run.pl JOB=1:$nj $data/split${nj}/JOB/check_tlg.log \
-    recognizer_main \
-        --wav_rspecifier=scp:$data/split${nj}/JOB/${aishell_wav_scp} \
-        --cmvn_file=$cmvn \
-        --model_path=$model_dir/avg_5.jit.pdmodel \
-        --streaming_chunk=30 \
-        --use_fbank=true \
-        --param_path=$model_dir/avg_5.jit.pdiparams \
-        --word_symbol_table=$wfst/words.txt \
-        --model_output_names=softmax_0.tmp_0,tmp_5,concat_0.tmp_0,concat_1.tmp_0 \
-        --model_cache_shapes="5-1-2048,5-1-2048" \
-        --graph_path=$wfst/TLG.fst --max_active=7500 \
-        --acoustic_scale=1.2 \
-        --result_wspecifier=ark,t:$data/split${nj}/JOB/result_check_tlg
-    cat $data/split${nj}/*/result_check_tlg > $exp/${label_file}_check_tlg
-    utils/compute-wer.py --char=1 --v=1 $text $exp/${label_file}_check_tlg > $exp/${wer}.check_tlg
-    echo "recognizer test have finished!!!"
-    echo "please checkout in ${exp}/${wer}.check_tlg"
-fi
-exit 0
--- a/speechx/examples/ds2_ol/aishell/local/split_data.sh
+++ b/speechx/examples/ds2_ol/aishell/local/split_data.sh
-#!/usr/bin/env bash
-set -eo pipefail
-data=$1
-scp=$2
-split_name=$3
-numsplit=$4
-# save in $data/split{n}
-# $scp to split
-# 
-if [[ ! $numsplit -gt 0 ]]; then
-  echo "Invalid num-split argument";
-  exit 1;
-fi
-directories=$(for n in `seq $numsplit`; do echo $data/split${numsplit}/$n; done)
-scp_splits=$(for n in `seq $numsplit`; do echo $data/split${numsplit}/$n/${split_name}; done)
-# if this mkdir fails due to argument-list being too long, iterate.
-if ! mkdir -p $directories >&/dev/null; then
-  for n in `seq $numsplit`; do
-    mkdir -p $data/split${numsplit}/$n
-  done
-fi
-echo "utils/split_scp.pl $scp $scp_splits"
-utils/split_scp.pl $scp $scp_splits
--- a/speechx/examples/ds2_ol/aishell/path.sh
+++ b/speechx/examples/ds2_ol/aishell/path.sh
-# This contains the locations of binarys build required for running the examples.
-MAIN_ROOT=`realpath $PWD/../../../../`
-SPEECHX_ROOT=$PWD/../../../
-SPEECHX_BUILD=$SPEECHX_ROOT/build/speechx
-SPEECHX_TOOLS=$SPEECHX_ROOT/tools
-TOOLS_BIN=$SPEECHX_TOOLS/valgrind/install/bin
-[ -d $SPEECHX_BUILD ] || { echo "Error: 'build/speechx' directory not found. please ensure that the project build successfully"; }
-export LC_AL=C
-# openfst bin & kaldi bin
-KALDI_DIR=$SPEECHX_ROOT/build/speechx/kaldi/
-OPENFST_DIR=$SPEECHX_ROOT/fc_patch/openfst-build/src
-# srilm
-export LIBLBFGS=${MAIN_ROOT}/tools/liblbfgs-1.10
-export LD_LIBRARY_PATH=${LD_LIBRARY_PATH:-}:${LIBLBFGS}/lib/.libs
-export SRILM=${MAIN_ROOT}/tools/srilm
-SPEECHX_BIN=$SPEECHX_BUILD/decoder:$SPEECHX_BUILD/frontend/audio
-export PATH=$PATH:$SPEECHX_BIN:$TOOLS_BIN:${SRILM}/bin:${SRILM}/bin/i686-m64:$KALDI_DIR/lmbin:$KALDI_DIR/fstbin:$OPENFST_DIR/bin
--- a/speechx/examples/ds2_ol/aishell/run.sh
+++ b/speechx/examples/ds2_ol/aishell/run.sh
-#!/bin/bash
-set -x
-set -e
-. path.sh
-nj=40
-stage=0
-stop_stage=100
-. utils/parse_options.sh
-# 1. compile
-if [ ! -d ${SPEECHX_BUILD} ]; then
-    pushd ${SPEECHX_ROOT} 
-    bash build.sh
-    popd
-fi
-# input
-mkdir -p data
-data=$PWD/data
-ckpt_dir=$data/model
-model_dir=$ckpt_dir/exp/deepspeech2_online/checkpoints/
-vocb_dir=$ckpt_dir/data/lang_char/
-# output
-mkdir -p exp
-exp=$PWD/exp
-aishell_wav_scp=aishell_test.scp
-if [ ${stage} -le 0 ] && [ ${stop_stage} -ge 0 ];then
-    if [ ! -d $data/test ]; then
-        # donwload dataset
-        pushd $data
-        wget -c https://paddlespeech.bj.bcebos.com/s2t/paddle_asr_online/aishell_test.zip
-        unzip  aishell_test.zip
-        popd
-        realpath $data/test/*/*.wav > $data/wavlist
-        awk -F '/' '{ print $(NF) }' $data/wavlist | awk -F '.' '{ print $1 }' > $data/utt_id
-        paste $data/utt_id $data/wavlist > $data/$aishell_wav_scp
-    fi
-    if [ ! -f $ckpt_dir/data/mean_std.json ]; then
-        # download model
-        mkdir -p $ckpt_dir
-        pushd $ckpt_dir
-        wget -c https://paddlespeech.bj.bcebos.com/s2t/aishell/asr0/asr0_deepspeech2_online_aishell_ckpt_0.2.0.model.tar.gz
-        tar xzfv asr0_deepspeech2_online_aishell_ckpt_0.2.0.model.tar.gz 
-        popd
-    fi
-    lm=$data/zh_giga.no_cna_cmn.prune01244.klm
-    if [ ! -f $lm ]; then
-        # download kenlm bin
-        pushd $data
-        wget -c https://deepspeech.bj.bcebos.com/zh_lm/zh_giga.no_cna_cmn.prune01244.klm
-        popd
-    fi
-fi
-# 3. make feature
-text=$data/test/text
-label_file=./aishell_result
-wer=./aishell_wer
-export GLOG_logtostderr=1
-cmvn=$data/cmvn.ark
-if [ ${stage} -le 1 ] && [ ${stop_stage} -ge 1 ]; then
-    # 3. convert cmvn format and compute linear feat
-    cmvn_json2kaldi_main --json_file=$ckpt_dir/data/mean_std.json --cmvn_write_path=$cmvn
-    ./local/split_data.sh $data $data/$aishell_wav_scp $aishell_wav_scp $nj
-    utils/run.pl JOB=1:$nj $data/split${nj}/JOB/feat.log \
-    compute_linear_spectrogram_main \
-        --wav_rspecifier=scp:$data/split${nj}/JOB/${aishell_wav_scp} \
-        --feature_wspecifier=ark,scp:$data/split${nj}/JOB/feat.ark,$data/split${nj}/JOB/feat.scp \
-        --cmvn_file=$cmvn \
-    echo "feature make have finished!!!"
-fi
-if [ ${stage} -le 2 ] && [ ${stop_stage} -ge 2 ]; then
-    #  decode w/o lm
-    utils/run.pl JOB=1:$nj $data/split${nj}/JOB/recog.wolm.log \
-    ctc_beam_search_decoder_main \
-        --feature_rspecifier=scp:$data/split${nj}/JOB/feat.scp \
-        --model_path=$model_dir/avg_1.jit.pdmodel \
-        --param_path=$model_dir/avg_1.jit.pdiparams \
-        --model_output_names=softmax_0.tmp_0,tmp_5,concat_0.tmp_0,concat_1.tmp_0 \
-	    --nnet_decoder_chunk=8 \
-        --dict_file=$vocb_dir/vocab.txt \
-        --result_wspecifier=ark,t:$data/split${nj}/JOB/result
-    cat $data/split${nj}/*/result > $exp/${label_file}
-    utils/compute-wer.py --char=1 --v=1 $text $exp/${label_file} > $exp/${wer}
-    echo "ctc-prefix-beam-search-decoder-ol without lm has finished!!!"
-    echo "please checkout in ${exp}/${wer}"
-    tail -n 7 $exp/${wer}
-fi
-if [ ${stage} -le 3 ] && [ ${stop_stage} -ge 3 ]; then
-    # decode w/ ngram lm with feature input
-    utils/run.pl JOB=1:$nj $data/split${nj}/JOB/recog.lm.log \
-    ctc_beam_search_decoder_main \
-        --feature_rspecifier=scp:$data/split${nj}/JOB/feat.scp \
-        --model_path=$model_dir/avg_1.jit.pdmodel \
-        --param_path=$model_dir/avg_1.jit.pdiparams \
-        --model_output_names=softmax_0.tmp_0,tmp_5,concat_0.tmp_0,concat_1.tmp_0 \
-	    --nnet_decoder_chunk=8 \
-        --dict_file=$vocb_dir/vocab.txt \
-        --lm_path=$lm \
-        --result_wspecifier=ark,t:$data/split${nj}/JOB/result_lm
-    cat $data/split${nj}/*/result_lm > $exp/${label_file}_lm
-    utils/compute-wer.py --char=1 --v=1 $text $exp/${label_file}_lm > $exp/${wer}.lm
-    echo "ctc-prefix-beam-search-decoder-ol with lm test has finished!!!"
-    echo "please checkout in ${exp}/${wer}.lm"
-    tail -n 7 $exp/${wer}.lm
-fi
-wfst=$data/wfst/
-if [ ${stage} -le 4 ] && [ ${stop_stage} -ge 4 ]; then
-    mkdir -p $wfst
-    if [ ! -f $wfst/aishell_graph.zip ]; then
-        # download TLG graph
-        pushd $wfst
-        wget -c https://paddlespeech.bj.bcebos.com/s2t/paddle_asr_online/aishell_graph.zip
-        unzip aishell_graph.zip
-        mv aishell_graph/* $wfst
-        popd
-    fi
-fi
-if [ ${stage} -le 4 ] && [ ${stop_stage} -ge 4 ]; then
-    #  decoder w/ TLG graph with feature input
-    utils/run.pl JOB=1:$nj $data/split${nj}/JOB/recog.wfst.log \
-    ctc_tlg_decoder_main \
-        --feature_rspecifier=scp:$data/split${nj}/JOB/feat.scp \
-        --model_path=$model_dir/avg_1.jit.pdmodel \
-        --param_path=$model_dir/avg_1.jit.pdiparams \
-        --word_symbol_table=$wfst/words.txt \
-        --model_output_names=softmax_0.tmp_0,tmp_5,concat_0.tmp_0,concat_1.tmp_0 \
-        --graph_path=$wfst/TLG.fst --max_active=7500 \
-	    --nnet_decoder_chunk=8 \
-        --acoustic_scale=1.2 \
-        --result_wspecifier=ark,t:$data/split${nj}/JOB/result_tlg
-    cat $data/split${nj}/*/result_tlg > $exp/${label_file}_tlg
-    utils/compute-wer.py --char=1 --v=1 $text $exp/${label_file}_tlg > $exp/${wer}.tlg
-    echo "wfst-decoder-ol have finished!!!"
-    echo "please checkout in ${exp}/${wer}.tlg"
-    tail -n 7 $exp/${wer}.tlg
-fi
-if [ ${stage} -le 5 ] && [ ${stop_stage} -ge 5 ]; then
-    #  recognize from wav file w/ TLG graph
-    utils/run.pl JOB=1:$nj $data/split${nj}/JOB/recognizer.log \
-    recognizer_main \
-        --wav_rspecifier=scp:$data/split${nj}/JOB/${aishell_wav_scp} \
-        --cmvn_file=$cmvn \
-        --model_path=$model_dir/avg_1.jit.pdmodel \
-        --param_path=$model_dir/avg_1.jit.pdiparams \
-        --word_symbol_table=$wfst/words.txt \
-	    --nnet_decoder_chunk=8 \
-        --model_output_names=softmax_0.tmp_0,tmp_5,concat_0.tmp_0,concat_1.tmp_0 \
-        --graph_path=$wfst/TLG.fst --max_active=7500 \
-        --acoustic_scale=1.2 \
-        --result_wspecifier=ark,t:$data/split${nj}/JOB/result_recognizer
-    cat $data/split${nj}/*/result_recognizer > $exp/${label_file}_recognizer
-    utils/compute-wer.py --char=1 --v=1 $text $exp/${label_file}_recognizer > $exp/${wer}.recognizer
-    echo "recognizer test have finished!!!"
-    echo "please checkout in ${exp}/${wer}.recognizer"
-    tail -n 7 $exp/${wer}.recognizer
-fi
\ No newline at end of file
--- a/speechx/examples/ds2_ol/aishell/run_fbank.sh
+++ b/speechx/examples/ds2_ol/aishell/run_fbank.sh
-#!/bin/bash
-set +x
-set -e
-. path.sh
-nj=40
-stage=0
-stop_stage=5
-. utils/parse_options.sh
-# 1. compile
-if [ ! -d ${SPEECHX_EXAMPLES} ]; then
-    pushd ${SPEECHX_ROOT} 
-    bash build.sh
-    popd
-fi
-# input
-mkdir -p data
-data=$PWD/data
-ckpt_dir=$data/fbank_model
-model_dir=$ckpt_dir/exp/deepspeech2_online/checkpoints/
-vocb_dir=$ckpt_dir/data/lang_char/
-# output
-mkdir -p exp
-exp=$PWD/exp
-lm=$data/zh_giga.no_cna_cmn.prune01244.klm
-aishell_wav_scp=aishell_test.scp
-if [ ${stage} -le 0 ] && [ ${stop_stage} -ge 0 ];then
-    if [ ! -d $data/test ]; then
-        pushd $data
-        wget -c https://paddlespeech.bj.bcebos.com/s2t/paddle_asr_online/aishell_test.zip
-        unzip  aishell_test.zip
-        popd
-        realpath $data/test/*/*.wav > $data/wavlist
-        awk -F '/' '{ print $(NF) }' $data/wavlist | awk -F '.' '{ print $1 }' > $data/utt_id
-        paste $data/utt_id $data/wavlist > $data/$aishell_wav_scp
-    fi
-    if [ ! -f $ckpt_dir/data/mean_std.json ]; then
-        mkdir -p $ckpt_dir
-        pushd $ckpt_dir
-        wget -c https://paddlespeech.bj.bcebos.com/s2t/wenetspeech/asr0/WIP1_asr0_deepspeech2_online_wenetspeech_ckpt_1.0.0a.model.tar.gz
-        tar xzfv WIP1_asr0_deepspeech2_online_wenetspeech_ckpt_1.0.0a.model.tar.gz
-        popd
-    fi
-    if [ ! -f $lm ]; then
-        pushd $data
-        wget -c https://deepspeech.bj.bcebos.com/zh_lm/zh_giga.no_cna_cmn.prune01244.klm
-        popd
-    fi
-fi
-# 3. make feature
-text=$data/test/text
-label_file=./aishell_result_fbank
-wer=./aishell_wer_fbank
-export GLOG_logtostderr=1
-cmvn=$data/cmvn_fbank.ark
-if [ ${stage} -le 1 ] && [ ${stop_stage} -ge 1 ]; then
-    # 3. convert cmvn format and compute fbank feat
-    cmvn_json2kaldi_main --json_file=$ckpt_dir/data/mean_std.json --cmvn_write_path=$cmvn --binary=false
-    ./local/split_data.sh $data $data/$aishell_wav_scp $aishell_wav_scp $nj
-    utils/run.pl JOB=1:$nj $data/split${nj}/JOB/feat.log \
-    compute_fbank_main \
-        --wav_rspecifier=scp:$data/split${nj}/JOB/${aishell_wav_scp} \
-        --feature_wspecifier=ark,scp:$data/split${nj}/JOB/fbank_feat.ark,$data/split${nj}/JOB/fbank_feat.scp \
-        --cmvn_file=$cmvn \
-        --streaming_chunk=36
-fi
-if [ ${stage} -le 2 ] && [ ${stop_stage} -ge 2 ]; then
-    #  decode w/ lm by feature
-    utils/run.pl JOB=1:$nj $data/split${nj}/JOB/recog.fbank.wolm.log \
-    ctc_beam_search_decoder_main \
-        --feature_rspecifier=scp:$data/split${nj}/JOB/fbank_feat.scp \
-        --model_path=$model_dir/avg_5.jit.pdmodel \
-        --param_path=$model_dir/avg_5.jit.pdiparams \
-        --model_output_names=softmax_0.tmp_0,tmp_5,concat_0.tmp_0,concat_1.tmp_0 \
-    	--model_cache_shapes="5-1-2048,5-1-2048" \
-	    --nnet_decoder_chunk=8 \
-        --dict_file=$vocb_dir/vocab.txt \
-        --result_wspecifier=ark,t:$data/split${nj}/JOB/result_fbank
-    cat $data/split${nj}/*/result_fbank > $exp/${label_file}
-    utils/compute-wer.py --char=1 --v=1 $text $exp/${label_file} > $exp/${wer}
-    tail -n 7 $exp/${wer}
-fi
-if [ ${stage} -le 3 ] && [ ${stop_stage} -ge 3 ]; then
-    #  decode with ngram lm by feature
-    utils/run.pl JOB=1:$nj $data/split${nj}/JOB/recog.fbank.lm.log \
-    ctc_beam_search_decoder_main \
-        --feature_rspecifier=scp:$data/split${nj}/JOB/fbank_feat.scp \
-        --model_path=$model_dir/avg_5.jit.pdmodel \
-        --param_path=$model_dir/avg_5.jit.pdiparams \
-        --model_output_names=softmax_0.tmp_0,tmp_5,concat_0.tmp_0,concat_1.tmp_0 \
-	    --model_cache_shapes="5-1-2048,5-1-2048" \
-	    --nnet_decoder_chunk=8 \
-        --dict_file=$vocb_dir/vocab.txt \
-        --lm_path=$lm \
-        --result_wspecifier=ark,t:$data/split${nj}/JOB/fbank_result_lm
-    cat $data/split${nj}/*/fbank_result_lm > $exp/${label_file}_lm
-    utils/compute-wer.py --char=1 --v=1 $text $exp/${label_file}_lm > $exp/${wer}.lm
-    tail -n 7 $exp/${wer}.lm
-fi
-wfst=$data/wfst_fbank/
-if [ ${stage} -le 4 ] && [ ${stop_stage} -ge 4 ]; then
-    mkdir -p $wfst
-    if [ ! -f $wfst/aishell_graph2.zip ]; then
-        pushd $wfst
-        wget -c https://paddlespeech.bj.bcebos.com/s2t/paddle_asr_online/aishell_graph2.zip
-        unzip aishell_graph2.zip
-        mv aishell_graph2/* $wfst
-        popd
-    fi
-fi
-if [ ${stage} -le 4 ] && [ ${stop_stage} -ge 4 ]; then
-    #  decode w/ TLG graph by feature
-    utils/run.pl JOB=1:$nj $data/split${nj}/JOB/recog.fbank.wfst.log \
-    ctc_tlg_decoder_main \
-        --feature_rspecifier=scp:$data/split${nj}/JOB/fbank_feat.scp \
-        --model_path=$model_dir/avg_5.jit.pdmodel \
-        --param_path=$model_dir/avg_5.jit.pdiparams \
-        --word_symbol_table=$wfst/words.txt \
-        --model_output_names=softmax_0.tmp_0,tmp_5,concat_0.tmp_0,concat_1.tmp_0 \
-	    --model_cache_shapes="5-1-2048,5-1-2048" \
-	    --nnet_decoder_chunk=8 \
-        --graph_path=$wfst/TLG.fst --max_active=7500 \
-        --acoustic_scale=1.2 \
-        --result_wspecifier=ark,t:$data/split${nj}/JOB/result_tlg
-    cat $data/split${nj}/*/result_tlg > $exp/${label_file}_tlg
-    utils/compute-wer.py --char=1 --v=1 $text $exp/${label_file}_tlg > $exp/${wer}.tlg
-    echo "wfst-decoder-ol have finished!!!"
-    echo "please checkout in ${exp}/${wer}.tlg"
-    tail -n 7 $exp/${wer}.tlg
-fi
-if [ ${stage} -le 5 ] && [ ${stop_stage} -ge 5 ]; then
-    # recgonize w/ TLG graph by wav
-    utils/run.pl JOB=1:$nj $data/split${nj}/JOB/fbank_recognizer.log \
-    recognizer_main \
-        --wav_rspecifier=scp:$data/split${nj}/JOB/${aishell_wav_scp} \
-        --cmvn_file=$cmvn \
-        --model_path=$model_dir/avg_5.jit.pdmodel \
-        --use_fbank=true \
-        --param_path=$model_dir/avg_5.jit.pdiparams \
-        --word_symbol_table=$wfst/words.txt \
-        --model_output_names=softmax_0.tmp_0,tmp_5,concat_0.tmp_0,concat_1.tmp_0 \
-        --model_cache_shapes="5-1-2048,5-1-2048" \
-	    --nnet_decoder_chunk=8 \
-        --graph_path=$wfst/TLG.fst --max_active=7500 \
-        --acoustic_scale=1.2 \
-        --result_wspecifier=ark,t:$data/split${nj}/JOB/result_fbank_recognizer
-    cat $data/split${nj}/*/result_fbank_recognizer > $exp/${label_file}_recognizer
-    utils/compute-wer.py --char=1 --v=1 $text $exp/${label_file}_recognizer > $exp/${wer}.recognizer
-    echo "recognizer test have finished!!!"
-    echo "please checkout in ${exp}/${wer}.recognizer"
-    tail -n 7 $exp/${wer}.recognizer
-fi
--- a/speechx/examples/ds2_ol/aishell/utils
+++ b/speechx/examples/ds2_ol/aishell/utils
-../../../../utils/
\ No newline at end of file
--- a/speechx/examples/ds2_ol/onnx/.gitignore
+++ b/speechx/examples/ds2_ol/onnx/.gitignore
-data
-log
-exp
--- a/speechx/examples/ds2_ol/onnx/README.md
+++ b/speechx/examples/ds2_ol/onnx/README.md
-# Convert DeepSpeech2 model to ONNX format
-> We recommend using U2/U2++ model instead of DS2, please see [here](../../u2pp_ol/wenetspeech/).
-This example demonstrate converting ds2 model to ONNX fromat.
-Please make sure [Paddle2ONNX](https://github.com/PaddlePaddle/Paddle2ONNX) and [onnx-simplifier](https://github.com/zh794390558/onnx-simplifier/tree/dyn_time_shape) version is correct.
-The example test with these packages installed:
-```
-paddle2onnx              0.9.8    # develop 62c5424e22cd93968dc831216fc9e0f0fce3d819
-paddleaudio              0.2.1
-paddlefsl                1.1.0
-paddlenlp                2.2.6
-paddlepaddle-gpu         2.2.2
-paddlespeech             0.0.0       # develop
-paddlespeech-ctcdecoders 0.2.0
-paddlespeech-feat        0.1.0
-onnx                     1.11.0
-onnx-simplifier          0.0.0       # https://github.com/zh794390558/onnx-simplifier/tree/dyn_time_shape
-onnxoptimizer            0.2.7
-onnxruntime              1.11.0
-```
-## Using
-```
-bash run.sh --stage 0 --stop_stage 5
-```
-1. convert deepspeech2 model to ONNX, using Paddle2ONNX.
-2. check paddleinference and onnxruntime output equal.
-3. optimize onnx model
-4. check paddleinference and optimized onnxruntime output equal.
-5. quantize onnx model
-6. check paddleinference and optimized onnxruntime output equal.
-For more details please see `run.sh`.
-## Outputs
-The optimized onnx model is `exp/model.opt.onnx`, quanted model is `exp/model.optset11.quant.onnx`.
-## [Results](https://github.com/PaddlePaddle/PaddleSpeech/wiki/ASR-Benchmark#streaming-asr)
-机器硬件：`CPU：Intel(R) Xeon(R) Gold 6271C CPU @ 2.60GHz`    
-测试脚本：`Streaming Server`      
-Acoustic Model | Model Size | enigne | dedoding_method | ctc_weight | decoding_chunk_size | num_decoding_left_chunk | RTF |
-|:-------------:| :-----: | :-----: | :------------:| :-----: | :-----: | :-----: |:-----:|
-| deepspeech2online_wenetspeech | 659MB | infernece | ctc_prefix_beam_search | - | 1 | - | 1.9108175171428279(utts=80) |
-| deepspeech2online_wenetspeech | 659MB | onnx | ctc_prefix_beam_search | - | 1 | - | 0.5617182449999291 (utts=80) |
-| deepspeech2online_wenetspeech | 166MB | onnx quant | ctc_prefix_beam_search | - | 1 | - | 0.44507715475808385 (utts=80) |
-> quant 和机器有关，不是所有机器都支持。ONNX quant测试机器指令集支持:
-> Flags:   fpu vme de pse tsc msr pae mce cx8 apic sep mtrr pge mca cmov pat pse36 clflush mmx fxsr sse sse2 ss ht syscall nx pdpe1gb rdtscp lm constant_tsc arch_perfmon rep_good nopl xtopology eagerfpu pni pclmulqdq ssse3 fma cx16 pcid sse4_1 sse4_2 x2apic movbe popcnt tsc_deadline_timer aes xsave avx f16c rdrand hypervisor lahf_lm abm 3dnowprefetch invpcid_single ssbd ibrs ibpb fsgsbase tsc_adjust bmi1 avx2 smep bmi2 erms invpcid mpx avx512f avx512dq rdseed adx smap clflushopt clwb avx512cd avx512bw avx512vl xsaveopt xsavec xgetbv1 arat umip pku ospke avx512_vnni spec_ctrl
--- a/speechx/examples/ds2_ol/onnx/local/infer_check.py
+++ b/speechx/examples/ds2_ol/onnx/local/infer_check.py
-#!/usr/bin/env python3
-# Copyright (c) 2022 PaddlePaddle Authors. All Rights Reserved.
-#
-# Licensed under the Apache License, Version 2.0 (the "License");
-# you may not use this file except in compliance with the License.
-# You may obtain a copy of the License at
-#
-#     http://www.apache.org/licenses/LICENSE-2.0
-#
-# Unless required by applicable law or agreed to in writing, software
-# distributed under the License is distributed on an "AS IS" BASIS,
-# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
-# See the License for the specific language governing permissions and
-# limitations under the License.
-import argparse
-import os
-import pickle
-import numpy as np
-import onnxruntime
-import paddle
-def parse_args():
-    parser = argparse.ArgumentParser(description=__doc__)
-    parser.add_argument(
-        '--input_file',
-        type=str,
-        default="static_ds2online_inputs.pickle",
-        help="aishell ds2 input data file. For wenetspeech, we only feed for infer model",
-    )
-    parser.add_argument(
-        '--model_type',
-        type=str,
-        default="aishell",
-        help="aishell(1024) or wenetspeech(2048)", )
-    parser.add_argument(
-        '--model_dir', type=str, default=".", help="paddle model dir.")
-    parser.add_argument(
-        '--model_prefix',
-        type=str,
-        default="avg_1.jit",
-        help="paddle model prefix.")
-    parser.add_argument(
-        '--onnx_model',
-        type=str,
-        default='./model.old.onnx',
-        help="onnx model.")
-    return parser.parse_args()
-if __name__ == '__main__':
-    FLAGS = parse_args()
-    # input and output
-    with open(FLAGS.input_file, 'rb') as f:
-        iodict = pickle.load(f)
-        print(iodict.keys())
-    audio_chunk = iodict['audio_chunk']
-    audio_chunk_lens = iodict['audio_chunk_lens']
-    chunk_state_h_box = iodict['chunk_state_h_box']
-    chunk_state_c_box = iodict['chunk_state_c_bos']
-    print("raw state shape: ", chunk_state_c_box.shape)
-    if FLAGS.model_type == 'wenetspeech':
-        chunk_state_h_box = np.repeat(chunk_state_h_box, 2, axis=-1)
-        chunk_state_c_box = np.repeat(chunk_state_c_box, 2, axis=-1)
-    print("state shape: ", chunk_state_c_box.shape)
-    # paddle
-    model = paddle.jit.load(os.path.join(FLAGS.model_dir, FLAGS.model_prefix))
-    res_chunk, res_lens, chunk_state_h, chunk_state_c = model(
-        paddle.to_tensor(audio_chunk),
-        paddle.to_tensor(audio_chunk_lens),
-        paddle.to_tensor(chunk_state_h_box),
-        paddle.to_tensor(chunk_state_c_box), )
-    # onnxruntime
-    options = onnxruntime.SessionOptions()
-    options.enable_profiling = True
-    sess = onnxruntime.InferenceSession(FLAGS.onnx_model, sess_options=options)
-    ort_res_chunk, ort_res_lens, ort_chunk_state_h, ort_chunk_state_c = sess.run(
-        ['softmax_0.tmp_0', 'tmp_5', 'concat_0.tmp_0', 'concat_1.tmp_0'], {
-            "audio_chunk": audio_chunk,
-            "audio_chunk_lens": audio_chunk_lens,
-            "chunk_state_h_box": chunk_state_h_box,
-            "chunk_state_c_box": chunk_state_c_box
-        })
-    print(sess.end_profiling())
-    # assert paddle equal ort
-    print(np.allclose(ort_res_chunk, res_chunk, atol=1e-6))
-    print(np.allclose(ort_res_lens, res_lens, atol=1e-6))
-    if FLAGS.model_type == 'aishell':
-        print(np.allclose(ort_chunk_state_h, chunk_state_h, atol=1e-6))
-        print(np.allclose(ort_chunk_state_c, chunk_state_c, atol=1e-6))
--- a/speechx/examples/ds2_ol/onnx/local/netron.sh
+++ b/speechx/examples/ds2_ol/onnx/local/netron.sh
-#!/bin/bash
-# show model 
-if [ $# != 1 ];then
-    echo "usage: $0 model_path"
-    exit 1
-fi
-file=$1
-pip install netron
-netron -p 8082 --host $(hostname -i) $file
\ No newline at end of file
--- a/speechx/examples/ds2_ol/onnx/local/onnx_clone.sh
+++ b/speechx/examples/ds2_ol/onnx/local/onnx_clone.sh
-#!/bin/bash
-# clone onnx repos
-git clone https://github.com/onnx/onnx.git
-git clone https://github.com/microsoft/onnxruntime.git
-git clone https://github.com/PaddlePaddle/Paddle2ONNX.git
\ No newline at end of file
--- a/speechx/examples/ds2_ol/onnx/local/onnx_convert_opset.py
+++ b/speechx/examples/ds2_ol/onnx/local/onnx_convert_opset.py
-#!/usr/bin/env python3
-import argparse
-import onnx
-from onnx import version_converter
-if __name__ == '__main__':
-    parser = argparse.ArgumentParser(prog=__doc__)
-    parser.add_argument(
-        "--model-file", type=str, required=True, help='path/to/the/model.onnx.')
-    parser.add_argument(
-        "--save-model",
-        type=str,
-        required=True,
-        help='path/to/saved/model.onnx.')
-    # Models must be opset10 or higher to be quantized.
-    parser.add_argument(
-        "--target-opset", type=int, default=11, help='path/to/the/model.onnx.')
-    args = parser.parse_args()
-    print(f"to opset: {args.target_opset}")
-    # Preprocessing: load the model to be converted.
-    model_path = args.model_file
-    original_model = onnx.load(model_path)
-    # print('The model before conversion:\n{}'.format(original_model))
-    # A full list of supported adapters can be found here:
-    # https://github.com/onnx/onnx/blob/main/onnx/version_converter.py#L21
-    # Apply the version conversion on the original model
-    converted_model = version_converter.convert_version(original_model,
-                                                        args.target_opset)
-    # print('The model after conversion:\n{}'.format(converted_model))
-    onnx.save(converted_model, args.save_model)
--- a/speechx/examples/ds2_ol/onnx/local/onnx_infer_shape.py
+++ b/speechx/examples/ds2_ol/onnx/local/onnx_infer_shape.py
--- a/speechx/examples/ds2_ol/onnx/local/onnx_opt.sh
+++ b/speechx/examples/ds2_ol/onnx/local/onnx_opt.sh
-#!/bin/bash
-set -e
-if [ $# != 3 ];then
-    # ./local/onnx_opt.sh model.old.onnx model.opt.onnx  "audio_chunk:1,-1,161  audio_chunk_lens:1 chunk_state_c_box:5,1,1024 chunk_state_h_box:5,1,1024"                                                 
-    echo "usage: $0 onnx.model.in onnx.model.out input_shape "
-    exit 1
-fi
-# onnx optimizer
-pip install onnx-simplifier
-in=$1
-out=$2
-input_shape=$3
-check_n=3
-onnxsim $in $out $check_n --dynamic-input-shape  --input-shape $input_shape
\ No newline at end of file
--- a/speechx/examples/ds2_ol/onnx/local/onnx_prune_model.py
+++ b/speechx/examples/ds2_ol/onnx/local/onnx_prune_model.py
-#!/usr/bin/env python3 -W ignore::DeprecationWarning
-# prune model by output names
-import argparse
-import copy
-import sys
-import onnx
-def parse_arguments():
-    parser = argparse.ArgumentParser()
-    parser.add_argument(
-        '--model',
-        required=True,
-        help='Path of directory saved the input model.')
-    parser.add_argument(
-        '--output_names',
-        required=True,
-        nargs='+',
-        help='The outputs of pruned model.')
-    parser.add_argument(
-        '--save_file', required=True, help='Path to save the new onnx model.')
-    return parser.parse_args()
-if __name__ == '__main__':
-    args = parse_arguments()
-    if len(set(args.output_names)) < len(args.output_names):
-        print(
-            "[ERROR] There's dumplicate name in --output_names, which is not allowed."
-        )
-        sys.exit(-1)
-    model = onnx.load(args.model)
-    # collect all node outputs and graph output
-    output_tensor_names = set()
-    for node in model.graph.node:
-        for out in node.output:
-            # may contain model output
-            output_tensor_names.add(out)
-    # for out in model.graph.output:
-    #     output_tensor_names.add(out.name)
-    for output_name in args.output_names:
-        if output_name not in output_tensor_names:
-            print(
-                "[ERROR] Cannot find output tensor name '{}' in onnx model graph.".
-                format(output_name))
-            sys.exit(-1)
-    output_node_indices = set()  # has output names
-    output_to_node = dict()  # all node outputs
-    for i, node in enumerate(model.graph.node):
-        for out in node.output:
-            output_to_node[out] = i
-            if out in args.output_names:
-                output_node_indices.add(i)
-    # from outputs find all the ancestors
-    reserved_node_indices = copy.deepcopy(
-        output_node_indices)  # nodes need to keep
-    reserved_inputs = set()  # model input to keep
-    new_output_node_indices = copy.deepcopy(output_node_indices)
-    while True and len(new_output_node_indices) > 0:
-        output_node_indices = copy.deepcopy(new_output_node_indices)
-        new_output_node_indices = set()
-        for out_node_idx in output_node_indices:
-            # backtrace to parenet
-            for ipt in model.graph.node[out_node_idx].input:
-                if ipt in output_to_node:
-                    reserved_node_indices.add(output_to_node[ipt])
-                    new_output_node_indices.add(output_to_node[ipt])
-                else:
-                    reserved_inputs.add(ipt)
-    num_inputs = len(model.graph.input)
-    num_outputs = len(model.graph.output)
-    num_nodes = len(model.graph.node)
-    print(
-        f"old graph has {num_inputs} inputs, {num_outputs} outpus, {num_nodes} nodes"
-    )
-    print(f"{len(reserved_node_indices)} node to keep.")
-    # del node not to keep
-    for idx in range(num_nodes - 1, -1, -1):
-        if idx not in reserved_node_indices:
-            del model.graph.node[idx]
-    # del graph input not to keep
-    for idx in range(num_inputs - 1, -1, -1):
-        if model.graph.input[idx].name not in reserved_inputs:
-            del model.graph.input[idx]
-    # del old graph outputs
-    for i in range(num_outputs):
-        del model.graph.output[0]
-    # new graph output as user input
-    for out in args.output_names:
-        model.graph.output.extend([onnx.ValueInfoProto(name=out)])
-    # infer shape
-    try:
-        from onnx_infer_shape import SymbolicShapeInference
-        model = SymbolicShapeInference.infer_shapes(
-            model,
-            int_max=2**31 - 1,
-            auto_merge=True,
-            guess_output_rank=False,
-            verbose=1)
-    except Exception as e:
-        print(f"skip infer shape step: {e}")
-    # check onnx model
-    onnx.checker.check_model(model)
-    # save onnx model
-    onnx.save(model, args.save_file)
-    print("[Finished] The new model saved in {}.".format(args.save_file))
-    print("[DEBUG INFO] The inputs of new model: {}".format(
-        [x.name for x in model.graph.input]))
-    print("[DEBUG INFO] The outputs of new model: {}".format(
-        [x.name for x in model.graph.output]))
--- a/speechx/examples/ds2_ol/onnx/local/onnx_rename_model.py
+++ b/speechx/examples/ds2_ol/onnx/local/onnx_rename_model.py
-#!/usr/bin/env python3 -W ignore::DeprecationWarning
-# rename node to new names
-import argparse
-import sys
-import onnx
-def parse_arguments():
-    parser = argparse.ArgumentParser()
-    parser.add_argument(
-        '--model',
-        required=True,
-        help='Path of directory saved the input model.')
-    parser.add_argument(
-        '--origin_names',
-        required=True,
-        nargs='+',
-        help='The original name you want to modify.')
-    parser.add_argument(
-        '--new_names',
-        required=True,
-        nargs='+',
-        help='The new name you want change to, the number of new_names should be same with the number of origin_names'
-    )
-    parser.add_argument(
-        '--save_file', required=True, help='Path to save the new onnx model.')
-    return parser.parse_args()
-if __name__ == '__main__':
-    args = parse_arguments()
-    if len(set(args.origin_names)) < len(args.origin_names):
-        print(
-            "[ERROR] There's dumplicate name in --origin_names, which is not allowed."
-        )
-        sys.exit(-1)
-    if len(set(args.new_names)) < len(args.new_names):
-        print(
-            "[ERROR] There's dumplicate name in --new_names, which is not allowed."
-        )
-        sys.exit(-1)
-    if len(args.new_names) != len(args.origin_names):
-        print(
-            "[ERROR] Number of --new_names must be same with the number of --origin_names."
-        )
-        sys.exit(-1)
-    model = onnx.load(args.model)
-    # collect input and all node output
-    output_tensor_names = set()
-    for ipt in model.graph.input:
-        output_tensor_names.add(ipt.name)
-    for node in model.graph.node:
-        for out in node.output:
-            output_tensor_names.add(out)
-    for origin_name in args.origin_names:
-        if origin_name not in output_tensor_names:
-            print(
-                f"[ERROR] Cannot find tensor name '{origin_name}' in onnx model graph."
-            )
-            sys.exit(-1)
-    for new_name in args.new_names:
-        if new_name in output_tensor_names:
-            print(
-                "[ERROR] The defined new_name '{}' is already exist in the onnx model, which is not allowed."
-            )
-            sys.exit(-1)
-    # rename graph input
-    for i, ipt in enumerate(model.graph.input):
-        if ipt.name in args.origin_names:
-            idx = args.origin_names.index(ipt.name)
-            model.graph.input[i].name = args.new_names[idx]
-    # rename node input and output
-    for i, node in enumerate(model.graph.node):
-        for j, ipt in enumerate(node.input):
-            if ipt in args.origin_names:
-                idx = args.origin_names.index(ipt)
-                model.graph.node[i].input[j] = args.new_names[idx]
-        for j, out in enumerate(node.output):
-            if out in args.origin_names:
-                idx = args.origin_names.index(out)
-                model.graph.node[i].output[j] = args.new_names[idx]
-    # rename graph output
-    for i, out in enumerate(model.graph.output):
-        if out.name in args.origin_names:
-            idx = args.origin_names.index(out.name)
-            model.graph.output[i].name = args.new_names[idx]
-    # check onnx model
-    onnx.checker.check_model(model)
-    # save model
-    onnx.save(model, args.save_file)
-    print("[Finished] The new model saved in {}.".format(args.save_file))
-    print("[DEBUG INFO] The inputs of new model: {}".format(
-        [x.name for x in model.graph.input]))
-    print("[DEBUG INFO] The outputs of new model: {}".format(
-        [x.name for x in model.graph.output]))
--- a/speechx/examples/ds2_ol/onnx/local/ort_dyanmic_quant.py
+++ b/speechx/examples/ds2_ol/onnx/local/ort_dyanmic_quant.py
-#!/usr/bin/env python3
-import argparse
-from onnxruntime.quantization import quantize_dynamic
-from onnxruntime.quantization import QuantType
-def quantize_onnx_model(onnx_model_path,
-                        quantized_model_path,
-                        nodes_to_exclude=[]):
-    print("Starting quantization...")
-    quantize_dynamic(
-        onnx_model_path,
-        quantized_model_path,
-        weight_type=QuantType.QInt8,
-        nodes_to_exclude=nodes_to_exclude)
-    print(f"Quantized model saved to: {quantized_model_path}")
-def main():
-    parser = argparse.ArgumentParser()
-    parser.add_argument(
-        "--model-in",
-        type=str,
-        required=True,
-        help="ONNX model", )
-    parser.add_argument(
-        "--model-out",
-        type=str,
-        required=True,
-        default='model.quant.onnx',
-        help="ONNX model", )
-    parser.add_argument(
-        "--nodes-to-exclude",
-        type=str,
-        required=True,
-        help="nodes to exclude. e.g. conv,linear.", )
-    args = parser.parse_args()
-    nodes_to_exclude = args.nodes_to_exclude.split(',')
-    quantize_onnx_model(args.model_in, args.model_out, nodes_to_exclude)
-if __name__ == "__main__":
-    main()
--- a/speechx/examples/ds2_ol/onnx/local/ort_opt.py
+++ b/speechx/examples/ds2_ol/onnx/local/ort_opt.py
-#!/usr/bin/env python3
-import argparse
-import onnxruntime as ort
-# onnxruntime optimizer.
-# https://onnxruntime.ai/docs/performance/graph-optimizations.html
-# https://onnxruntime.ai/docs/api/python/api_summary.html#api
-def parse_arguments():
-    parser = argparse.ArgumentParser()
-    parser.add_argument(
-        '--model_in', required=True, type=str, help='Path to onnx model.')
-    parser.add_argument(
-        '--opt_level',
-        required=True,
-        type=int,
-        default=0,
-        choices=[0, 1, 2],
-        help='Path to onnx model.')
-    parser.add_argument(
-        '--model_out', required=True, help='path to save the optimized model.')
-    parser.add_argument('--debug', default=False, help='output debug info.')
-    return parser.parse_args()
-if __name__ == '__main__':
-    args = parse_arguments()
-    sess_options = ort.SessionOptions()
-    # Set graph optimization level
-    print(f"opt level: {args.opt_level}")
-    if args.opt_level == 0:
-        sess_options.graph_optimization_level = ort.GraphOptimizationLevel.ORT_ENABLE_BASIC
-    elif args.opt_level == 1:
-        sess_options.graph_optimization_level = ort.GraphOptimizationLevel.ORT_ENABLE_EXTENDED
-    else:
-        sess_options.graph_optimization_level = ort.GraphOptimizationLevel.ORT_ENABLE_ALL
-    # To enable model serialization after graph optimization set this
-    sess_options.optimized_model_filepath = args.model_out
-    session = ort.InferenceSession(args.model_in, sess_options)
--- a/speechx/examples/ds2_ol/onnx/local/tonnx.sh
+++ b/speechx/examples/ds2_ol/onnx/local/tonnx.sh
-#!/bin/bash
-if [ $# != 4 ];then
-    # local/tonnx.sh data/exp/deepspeech2_online/checkpoints avg_1.jit.pdmodel avg_1.jit.pdiparams exp/model.onnx   
-    echo "usage: $0 model_dir model_name param_name onnx_output_name"
-    exit 1
-fi
-dir=$1
-model=$2
-param=$3
-output=$4
-pip install paddle2onnx
-pip install onnx
-# https://github.com/PaddlePaddle/Paddle2ONNX#%E5%91%BD%E4%BB%A4%E8%A1%8C%E8%BD%AC%E6%8D%A2
- # opset10 support quantize
-paddle2onnx --model_dir $dir \
-            --model_filename $model \
-            --params_filename $param \
-            --save_file $output \
-            --enable_dev_version True \
-            --opset_version 11 \
-            --enable_onnx_checker True
\ No newline at end of file
--- a/speechx/examples/ds2_ol/onnx/path.sh
+++ b/speechx/examples/ds2_ol/onnx/path.sh
-# This contains the locations of binarys build required for running the examples.
-MAIN_ROOT=`realpath $PWD/../../../../`
-SPEECHX_ROOT=$PWD/../../../
-SPEECHX_BUILD=$SPEECHX_ROOT/build/speechx
-SPEECHX_TOOLS=$SPEECHX_ROOT/tools
-TOOLS_BIN=$SPEECHX_TOOLS/valgrind/install/bin
-[ -d $SPEECHX_BUILD ] || { echo "Error: 'build/speechx' directory not found. please ensure that the project build successfully"; }
-export LC_AL=C
-export PATH=$PATH:$TOOLS_BIN
--- a/speechx/examples/ds2_ol/onnx/run.sh
+++ b/speechx/examples/ds2_ol/onnx/run.sh
-#!/bin/bash
-set -e
-. path.sh
-stage=0
-stop_stage=50
-tarfile=asr0_deepspeech2_online_wenetspeech_ckpt_1.0.2.model.tar.gz
-#tarfile=asr0_deepspeech2_online_aishell_fbank161_ckpt_1.0.1.model.tar.gz
-model_prefix=avg_10.jit
-#model_prefix=avg_1.jit
-model=${model_prefix}.pdmodel
-param=${model_prefix}.pdiparams
-. utils/parse_options.sh
-data=data
-exp=exp
-mkdir -p $data $exp
-dir=$data/exp/deepspeech2_online/checkpoints
-# wenetspeech or aishell
-model_type=$(echo $tarfile | cut -d '_' -f 4)
-if [ ${stage} -le 0 ] && [ ${stop_stage} -ge 0 ];then
-    test -f $data/$tarfile || wget -P $data -c https://paddlespeech.bj.bcebos.com/s2t/$model_type/asr0/$tarfile
-    # wenetspeech ds2 model
-    pushd $data
-    tar zxvf $tarfile 
-    popd
-    # ds2 model demo inputs
-    pushd $exp
-    wget -c http://paddlespeech.bj.bcebos.com/speechx/examples/ds2_ol/onnx/static_ds2online_inputs.pickle
-    popd
-fi
-input_file=$exp/static_ds2online_inputs.pickle
-test -e $input_file
-if [ ${stage} -le 1 ] && [ ${stop_stage} -ge 1 ];then
-    # to onnx
-   ./local/tonnx.sh $dir $model $param $exp/model.onnx
-   ./local/infer_check.py --input_file $input_file --model_type $model_type --model_dir $dir --model_prefix $model_prefix --onnx_model $exp/model.onnx
-fi
-if [ ${stage} -le 2 ] && [ ${stop_stage} -ge 2 ] ;then
-    # ort graph optmize
-    ./local/ort_opt.py --model_in $exp/model.onnx --opt_level 0 --model_out $exp/model.ort.opt.onnx
-    ./local/infer_check.py --input_file $input_file --model_type $model_type  --model_dir $dir --model_prefix $model_prefix --onnx_model $exp/model.ort.opt.onnx
-fi
-if [ ${stage} -le 3 ] && [ ${stop_stage} -ge 3 ];then
-    # convert opset_num to 11
-    ./local/onnx_convert_opset.py --target-opset 11  --model-file $exp/model.ort.opt.onnx --save-model $exp/model.optset11.onnx 
-    # quant model
-    nodes_to_exclude='p2o.Conv.0,p2o.Conv.2'
-    ./local/ort_dyanmic_quant.py --model-in $exp/model.optset11.onnx --model-out $exp/model.optset11.quant.onnx --nodes-to-exclude "${nodes_to_exclude}"
-    ./local/infer_check.py --input_file $input_file --model_type $model_type  --model_dir $dir --model_prefix $model_prefix --onnx_model $exp/model.optset11.quant.onnx
-fi
-# aishell rnn hidden is 1024
-# wenetspeech rnn hiddn is 2048
-if [ $model_type == 'aishell' ];then
-    input_shape="audio_chunk:1,-1,161  audio_chunk_lens:1 chunk_state_c_box:5,1,1024 chunk_state_h_box:5,1,1024"  
-elif [ $model_type == 'wenetspeech' ];then
-    input_shape="audio_chunk:1,-1,161  audio_chunk_lens:1 chunk_state_c_box:5,1,2048 chunk_state_h_box:5,1,2048"  
-else
-    echo "not support: $model_type"
-    exit -1
-fi
-if [ ${stage} -le 51 ] && [ ${stop_stage} -ge 51 ] ;then
-    # wenetspeech ds2 model execed 2GB limit, will error.
-    # simplifying onnx model
-    ./local/onnx_opt.sh $exp/model.onnx $exp/model.opt.onnx  "$input_shape"
-    ./local/infer_check.py --input_file $input_file --model_type $model_type  --model_dir $dir --model_prefix $model_prefix --onnx_model $exp/model.opt.onnx
-fi
--- a/speechx/examples/ds2_ol/onnx/utils
+++ b/speechx/examples/ds2_ol/onnx/utils
-../../../../utils/
\ No newline at end of file
--- a/speechx/examples/ds2_ol/websocket/.gitignore
+++ b/speechx/examples/ds2_ol/websocket/.gitignore
-data
-exp
--- a/speechx/examples/ds2_ol/websocket/README.md
+++ b/speechx/examples/ds2_ol/websocket/README.md
-#  Streaming DeepSpeech2 Server with WebSocket
-This example is about using `websocket` as streaming deepspeech2 server. For deepspeech2 model training please see [here](../../../../examples/aishell/asr0/).
-The websocket protocal is same to [PaddleSpeech Server](../../../../demos/streaming_asr_server/), 
-for detail of implementation please see [here](../../../speechx/protocol/websocket/).
-## Source path.sh
-```bash
-. path.sh
-```
-SpeechX bins is under `echo $SPEECHX_BUILD`, more info please see `path.sh`.
-## Start WebSocket Server
-```bash
-bash websoket_server.sh
-```
-The output is like below:
-```text
-I1130 02:19:32.029882 12856 cmvn_json2kaldi_main.cc:39] cmvn josn path: /workspace/zhanghui/PaddleSpeech/speechx/examples/ds2_ol/websocket/data/model/data/mean_std.json
-I1130 02:19:32.032230 12856 cmvn_json2kaldi_main.cc:73] nframe: 907497
-I1130 02:19:32.032564 12856 cmvn_json2kaldi_main.cc:85] cmvn stats have write into: /workspace/zhanghui/PaddleSpeech/speechx/examples/ds2_ol/websocket/data/cmvn.ark
-I1130 02:19:32.032579 12856 cmvn_json2kaldi_main.cc:86] Binary: 1
-I1130 02:19:32.798342 12937 feature_pipeline.h:53] cmvn file: /workspace/zhanghui/PaddleSpeech/speechx/examples/ds2_ol/websocket/data/cmvn.ark
-I1130 02:19:32.798542 12937 feature_pipeline.h:58] dither: 0
-I1130 02:19:32.798583 12937 feature_pipeline.h:60] frame shift ms: 10
-I1130 02:19:32.798588 12937 feature_pipeline.h:62] feature type: linear
-I1130 02:19:32.798596 12937 feature_pipeline.h:80] frame length ms: 20
-I1130 02:19:32.798601 12937 feature_pipeline.h:88] subsampling rate: 4
-I1130 02:19:32.798606 12937 feature_pipeline.h:90] nnet receptive filed length: 7
-I1130 02:19:32.798611 12937 feature_pipeline.h:92] nnet chunk size: 1
-I1130 02:19:32.798615 12937 feature_pipeline.h:94] frontend fill zeros: 0
-I1130 02:19:32.798630 12937 nnet_itf.h:52] subsampling rate: 4
-I1130 02:19:32.798635 12937 nnet_itf.h:54] model path: /workspace/zhanghui/PaddleSpeech/speechx/examples/ds2_ol/websocket/data/model/exp/deepspeech2_online/checkpoints//avg_1.jit.pdmodel
-I1130 02:19:32.798640 12937 nnet_itf.h:57] param path: /workspace/zhanghui/PaddleSpeech/speechx/examples/ds2_ol/websocket/data/model/exp/deepspeech2_online/checkpoints//avg_1.jit.pdiparams
-I1130 02:19:32.798643 12937 nnet_itf.h:59] DS2 param: 
-I1130 02:19:32.798647 12937 nnet_itf.h:61]   cache names: chunk_state_h_box,chunk_state_c_box
-I1130 02:19:32.798652 12937 nnet_itf.h:63]   cache shape: 5-1-1024,5-1-1024
-I1130 02:19:32.798656 12937 nnet_itf.h:65]   input names: audio_chunk,audio_chunk_lens,chunk_state_h_box,chunk_state_c_box
-I1130 02:19:32.798660 12937 nnet_itf.h:67]   output names: softmax_0.tmp_0,tmp_5,concat_0.tmp_0,concat_1.tmp_0
-I1130 02:19:32.798664 12937 ctc_tlg_decoder.h:41] fst path: /workspace/zhanghui/PaddleSpeech/speechx/examples/ds2_ol/websocket/data/wfst//TLG.fst
-I1130 02:19:32.798669 12937 ctc_tlg_decoder.h:42] fst symbole table: /workspace/zhanghui/PaddleSpeech/speechx/examples/ds2_ol/websocket/data/wfst//words.txt
-I1130 02:19:32.798673 12937 ctc_tlg_decoder.h:47] LatticeFasterDecoder max active: 7500
-I1130 02:19:32.798677 12937 ctc_tlg_decoder.h:49] LatticeFasterDecoder beam: 15
-I1130 02:19:32.798681 12937 ctc_tlg_decoder.h:50] LatticeFasterDecoder lattice_beam: 7.5
-I1130 02:19:32.798708 12937 websocket_server_main.cc:37] Listening at port 8082
-```
-## Start WebSocket Client
-```bash
-bash websocket_client.sh
-```
-This script using AISHELL-1 test data to call websocket server.
-The input is specific by `--wav_rspecifier=scp:$data/$aishell_wav_scp`.
-The `scp` file which look like this:
-```text
-# head data/split1/1/aishell_test.scp 
-BAC009S0764W0121        /workspace/PaddleSpeech/speechx/examples/u2pp_ol/wenetspeech/data/test/S0764/BAC009S0764W0121.wav
-BAC009S0764W0122        /workspace/PaddleSpeech/speechx/examples/u2pp_ol/wenetspeech/data/test/S0764/BAC009S0764W0122.wav
-...
-BAC009S0764W0125        /workspace/PaddleSpeech/speechx/examples/u2pp_ol/wenetspeech/data/test/S0764/BAC009S0764W0125.wav
-```
-If you want to recognize one wav, you can make `scp` file like this:
-```text
-key  path/to/wav/file
-```
--- a/speechx/examples/ds2_ol/websocket/path.sh
+++ b/speechx/examples/ds2_ol/websocket/path.sh
-# This contains the locations of binarys build required for running the examples.
-SPEECHX_ROOT=$PWD/../../../
-SPEECHX_BUILD=$SPEECHX_ROOT/build/speechx
-SPEECHX_TOOLS=$SPEECHX_ROOT/tools
-TOOLS_BIN=$SPEECHX_TOOLS/valgrind/install/bin
-[ -d $SPEECHX_BUILD ] || { echo "Error: 'build/speechx' directory not found. please ensure that the project build successfully"; }
-export LC_AL=C
-SPEECHX_BIN=$SPEECHX_BUILD/protocol/websocket:$SPEECHX_BUILD/frontend/audio
-export PATH=$PATH:$SPEECHX_BIN:$TOOLS_BIN
--- a/speechx/examples/ds2_ol/websocket/websocket_client.sh
+++ b/speechx/examples/ds2_ol/websocket/websocket_client.sh
-#!/bin/bash
-set +x
-set -e
-. path.sh
-# 1. compile
-if [ ! -d ${SPEECHX_EXAMPLES} ]; then
-    pushd ${SPEECHX_ROOT} 
-    bash build.sh
-    popd
-fi
-# input
-mkdir -p data
-data=$PWD/data
-# output
-aishell_wav_scp=aishell_test.scp
-if [ ! -d $data/test ]; then
-    pushd $data
-    wget -c https://paddlespeech.bj.bcebos.com/s2t/paddle_asr_online/aishell_test.zip
-    unzip  aishell_test.zip
-    popd
-    realpath $data/test/*/*.wav > $data/wavlist
-    awk -F '/' '{ print $(NF) }' $data/wavlist | awk -F '.' '{ print $1 }' > $data/utt_id
-    paste $data/utt_id $data/wavlist > $data/$aishell_wav_scp
-fi
-export GLOG_logtostderr=1
-# websocket client
-websocket_client_main \
-    --wav_rspecifier=scp:$data/$aishell_wav_scp --streaming_chunk=0.5
--- a/speechx/examples/ds2_ol/websocket/websocket_server.sh
+++ b/speechx/examples/ds2_ol/websocket/websocket_server.sh
-#!/bin/bash
-set +x
-set -e
-. path.sh
-# 1. compile
-if [ ! -d ${SPEECHX_EXAMPLES} ]; then
-    pushd ${SPEECHX_ROOT} 
-    bash build.sh
-    popd
-fi
-# input
-mkdir -p data
-data=$PWD/data
-ckpt_dir=$data/model
-model_dir=$ckpt_dir/exp/deepspeech2_online/checkpoints/
-vocb_dir=$ckpt_dir/data/lang_char/
-if [ ! -f $ckpt_dir/data/mean_std.json ]; then
-    mkdir -p $ckpt_dir
-    pushd $ckpt_dir
-    wget -c https://paddlespeech.bj.bcebos.com/s2t/aishell/asr0/asr0_deepspeech2_online_aishell_ckpt_0.2.0.model.tar.gz
-    tar xzfv asr0_deepspeech2_online_aishell_ckpt_0.2.0.model.tar.gz 
-    popd
-fi
-export GLOG_logtostderr=1
-# 3. gen cmvn 
-cmvn=$data/cmvn.ark
-cmvn_json2kaldi_main --json_file=$ckpt_dir/data/mean_std.json --cmvn_write_path=$cmvn
-wfst=$data/wfst/
-mkdir -p $wfst
-if [ ! -f $wfst/aishell_graph.zip ]; then
-    pushd $wfst
-    wget -c https://paddlespeech.bj.bcebos.com/s2t/paddle_asr_online/aishell_graph.zip
-    unzip aishell_graph.zip
-    mv aishell_graph/* $wfst
-    popd
-fi
-# 5. test websocket server 
-websocket_server_main \
-    --cmvn_file=$cmvn \
-    --model_path=$model_dir/avg_1.jit.pdmodel \
-    --param_path=$model_dir/avg_1.jit.pdiparams \
-    --word_symbol_table=$wfst/words.txt \
-    --model_output_names=softmax_0.tmp_0,tmp_5,concat_0.tmp_0,concat_1.tmp_0 \
-    --graph_path=$wfst/TLG.fst --max_active=7500 \
-    --acoustic_scale=1.2 
--- a/speechx/examples/u2pp_ol/wenetspeech/path.sh
+++ b/speechx/examples/u2pp_ol/wenetspeech/path.sh
@@ -3,7 +3,7 @@
 unset GREP_OPTIONS
 SPEECHX_ROOT=$PWD/../../../
-SPEECHX_BUILD=$SPEECHX_ROOT/build/speechx
+SPEECHX_BUILD=$SPEECHX_ROOT/build/speechx/asr
 SPEECHX_TOOLS=$SPEECHX_ROOT/tools
 TOOLS_BIN=$SPEECHX_TOOLS/valgrind/install/bin
@@ -12,7 +12,7 @@ TOOLS_BIN=$SPEECHX_TOOLS/valgrind/install/bin
 export LC_AL=C
-export PATH=$PATH:$TOOLS_BIN:$SPEECHX_BUILD/nnet:$SPEECHX_BUILD/decoder:$SPEECHX_BUILD/frontend/audio:$SPEECHX_BUILD/recognizer
+export PATH=$PATH:$TOOLS_BIN:$SPEECHX_BUILD/nnet:$SPEECHX_BUILD/decoder:$SPEECHX_BUILD/../common/frontend/audio:$SPEECHX_BUILD/recognizer
 PADDLE_LIB_PATH=$(python -c "import os; import paddle; include_dir=paddle.sysconfig.get_include(); paddle_dir=os.path.split(include_dir)[0]; libs_dir=os.path.join(paddle_dir, 'libs'); fluid_dir=os.path.join(paddle_dir, 'fluid'); out=':'.join([libs_dir, fluid_dir]); print(out);")
 export LD_LIBRARY_PATH=$PADDLE_LIB_PATH:$LD_LIBRARY_PATH
--- a/speechx/speechx/asr/decoder/CMakeLists.txt
+++ b/speechx/speechx/asr/decoder/CMakeLists.txt
-include_directories(${CMAKE_CURRENT_SOURCE_DIR/ctc_decoders})
 set(srcs)
-if (USING_DS2)
 list(APPEND srcs
-  ctc_decoders/decoder_utils.cpp
+  ctc_prefix_beam_search_decoder.cc
-  ctc_decoders/path_trie.cpp
-  ctc_decoders/scorer.cpp
-  ctc_beam_search_decoder.cc
-  ctc_tlg_decoder.cc
 )
-endif()
-if (USING_U2)
-  list(APPEND srcs
-    ctc_prefix_beam_search_decoder.cc
-  )
-endif()
 add_library(decoder STATIC ${srcs})
-target_link_libraries(decoder PUBLIC kenlm utils fst frontend nnet kaldi-decoder)
+target_link_libraries(decoder PUBLIC utils fst frontend nnet kaldi-decoder)
 # test
-if (USING_DS2)
+set(TEST_BINS 
-  set(BINS 
+  ctc_prefix_beam_search_decoder_main
-    ctc_beam_search_decoder_main
+)
-    nnet_logprob_decoder_main
-    ctc_tlg_decoder_main
-  )
-  foreach(bin_name IN LISTS BINS)
-    add_executable(${bin_name} ${CMAKE_CURRENT_SOURCE_DIR}/${bin_name}.cc)
-    target_include_directories(${bin_name} PRIVATE ${SPEECHX_ROOT} ${SPEECHX_ROOT}/kaldi)
-    target_link_libraries(${bin_name} PUBLIC nnet decoder fst utils gflags glog kaldi-base kaldi-matrix kaldi-util ${DEPS})
-  endforeach()
-endif()
-if (USING_U2)
-  set(TEST_BINS 
-    ctc_prefix_beam_search_decoder_main
-  )
-  foreach(bin_name IN LISTS TEST_BINS)
-    add_executable(${bin_name} ${CMAKE_CURRENT_SOURCE_DIR}/${bin_name}.cc)
-    target_include_directories(${bin_name} PRIVATE ${SPEECHX_ROOT} ${SPEECHX_ROOT}/kaldi)
-    target_link_libraries(${bin_name} nnet decoder fst utils gflags glog kaldi-base kaldi-matrix kaldi-util)
-    target_compile_options(${bin_name}  PRIVATE ${PADDLE_COMPILE_FLAGS})
-    target_include_directories(${bin_name}  PRIVATE ${pybind11_INCLUDE_DIRS} ${PROJECT_SOURCE_DIR})
-    target_link_libraries(${bin_name}  ${PYTHON_LIBRARIES} ${PADDLE_LINK_FLAGS})
-  endforeach()
-endif()
+foreach(bin_name IN LISTS TEST_BINS)
+  add_executable(${bin_name} ${CMAKE_CURRENT_SOURCE_DIR}/${bin_name}.cc)
+  target_include_directories(${bin_name} PRIVATE ${SPEECHX_ROOT} ${SPEECHX_ROOT}/kaldi)
+  target_link_libraries(${bin_name} nnet decoder fst utils gflags glog kaldi-base kaldi-matrix kaldi-util)
+  target_compile_options(${bin_name}  PRIVATE ${PADDLE_COMPILE_FLAGS})
+  target_include_directories(${bin_name}  PRIVATE ${pybind11_INCLUDE_DIRS} ${PROJECT_SOURCE_DIR})
+  target_link_libraries(${bin_name}  ${PYTHON_LIBRARIES} ${PADDLE_LINK_FLAGS})
+endforeach()
--- a/speechx/speechx/asr/decoder/ctc_beam_search_decoder.cc
+++ b/speechx/speechx/asr/decoder/ctc_beam_search_decoder.cc
-// Copyright (c) 2022 PaddlePaddle Authors. All Rights Reserved.
-//
-// Licensed under the Apache License, Version 2.0 (the "License");
-// you may not use this file except in compliance with the License.
-// You may obtain a copy of the License at
-//
-//     http://www.apache.org/licenses/LICENSE-2.0
-//
-// Unless required by applicable law or agreed to in writing, software
-// distributed under the License is distributed on an "AS IS" BASIS,
-// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
-// See the License for the specific language governing permissions and
-// limitations under the License.
-#include "decoder/ctc_beam_search_decoder.h"
-#include "base/common.h"
-#include "decoder/ctc_decoders/decoder_utils.h"
-#include "utils/file_utils.h"
-namespace ppspeech {
-using std::vector;
-using FSTMATCH = fst::SortedMatcher<fst::StdVectorFst>;
-CTCBeamSearch::CTCBeamSearch(const CTCBeamSearchOptions& opts)
-    : opts_(opts), init_ext_scorer_(nullptr), space_id_(-1), root_(nullptr) {
-    LOG(INFO) << "dict path: " << opts_.dict_file;
-    if (!ReadFileToVector(opts_.dict_file, &vocabulary_)) {
-        LOG(INFO) << "load the dict failed";
-    }
-    LOG(INFO) << "read the vocabulary success, dict size: "
-              << vocabulary_.size();
-    LOG(INFO) << "language model path: " << opts_.lm_path;
-    if (opts_.lm_path != "") {
-        init_ext_scorer_ = std::make_shared<Scorer>(
-            opts_.alpha, opts_.beta, opts_.lm_path, vocabulary_);
-    }
-    CHECK_EQ(opts_.blank, 0);
-    auto it = std::find(vocabulary_.begin(), vocabulary_.end(), " ");
-    space_id_ = it - vocabulary_.begin();
-    // if no space in vocabulary
-    if (static_cast<size_t>(space_id_) >= vocabulary_.size()) {
-        space_id_ = -2;
-    }
-}
-void CTCBeamSearch::Reset() {
-    // num_frame_decoded_ = 0;
-    // ResetPrefixes();
-    InitDecoder();
-}
-void CTCBeamSearch::InitDecoder() {
-    num_frame_decoded_ = 0;
-    // ResetPrefixes();
-    prefixes_.clear();
-    root_ = std::make_shared<PathTrie>();
-    root_->score = root_->log_prob_b_prev = 0.0;
-    prefixes_.push_back(root_.get());
-    if (init_ext_scorer_ != nullptr &&
-        !init_ext_scorer_->is_character_based()) {
-        auto fst_dict =
-            static_cast<fst::StdVectorFst*>(init_ext_scorer_->dictionary);
-        fst::StdVectorFst* dict_ptr = fst_dict->Copy(true);
-        root_->set_dictionary(dict_ptr);
-        auto matcher = std::make_shared<FSTMATCH>(*dict_ptr, fst::MATCH_INPUT);
-        root_->set_matcher(matcher);
-    }
-}
-void CTCBeamSearch::Decode(
-    std::shared_ptr<kaldi::DecodableInterface> decodable) {
-    return;
-}
-// todo rename, refactor
-void CTCBeamSearch::AdvanceDecode(
-    const std::shared_ptr<kaldi::DecodableInterface>& decodable) {
-    while (1) {
-        vector<vector<BaseFloat>> likelihood;
-        vector<BaseFloat> frame_prob;
-        bool flag = decodable->FrameLikelihood(num_frame_decoded_, &frame_prob);
-        if (flag == false) break;
-        likelihood.push_back(frame_prob);
-        AdvanceDecoding(likelihood);
-    }
-}
-void CTCBeamSearch::ResetPrefixes() {
-    for (size_t i = 0; i < prefixes_.size(); i++) {
-        if (prefixes_[i] != nullptr) {
-            delete prefixes_[i];
-            prefixes_[i] = nullptr;
-        }
-    }
-    prefixes_.clear();
-}
-int CTCBeamSearch::DecodeLikelihoods(const vector<vector<float>>& probs,
-                                     const vector<string>& nbest_words) {
-    kaldi::Timer timer;
-    AdvanceDecoding(probs);
-    LOG(INFO) << "ctc decoding elapsed time(s) "
-              << static_cast<float>(timer.Elapsed()) / 1000.0f;
-    return 0;
-}
-vector<std::pair<double, string>> CTCBeamSearch::GetNBestPath(int n) {
-    int beam_size = n == -1 ? opts_.beam_size : std::min(n, opts_.beam_size);
-    return get_beam_search_result(prefixes_, vocabulary_, beam_size);
-}
-vector<std::pair<double, string>> CTCBeamSearch::GetNBestPath() {
-    return GetNBestPath(-1);
-}
-string CTCBeamSearch::GetBestPath() {
-    std::vector<std::pair<double, std::string>> result;
-    result = get_beam_search_result(prefixes_, vocabulary_, opts_.beam_size);
-    return result[0].second;
-}
-string CTCBeamSearch::GetFinalBestPath() {
-    CalculateApproxScore();
-    LMRescore();
-    return GetBestPath();
-}
-void CTCBeamSearch::AdvanceDecoding(const vector<vector<BaseFloat>>& probs) {
-    size_t num_time_steps = probs.size();
-    size_t beam_size = opts_.beam_size;
-    double cutoff_prob = opts_.cutoff_prob;
-    size_t cutoff_top_n = opts_.cutoff_top_n;
-    vector<vector<double>> probs_seq(probs.size(),
-                                     vector<double>(probs[0].size(), 0));
-    int row = probs.size();
-    int col = probs[0].size();
-    for (int i = 0; i < row; i++) {
-        for (int j = 0; j < col; j++) {
-            probs_seq[i][j] = static_cast<double>(probs[i][j]);
-        }
-    }
-    for (size_t time_step = 0; time_step < num_time_steps; time_step++) {
-        const auto& prob = probs_seq[time_step];
-        float min_cutoff = -NUM_FLT_INF;
-        bool full_beam = false;
-        if (init_ext_scorer_ != nullptr) {
-            size_t num_prefixes_ = std::min(prefixes_.size(), beam_size);
-            std::sort(prefixes_.begin(),
-                      prefixes_.begin() + num_prefixes_,
-                      prefix_compare);
-            if (num_prefixes_ == 0) {
-                continue;
-            }
-            min_cutoff = prefixes_[num_prefixes_ - 1]->score +
-                         std::log(prob[opts_.blank]) -
-                         std::max(0.0, init_ext_scorer_->beta);
-            full_beam = (num_prefixes_ == beam_size);
-        }
-        vector<std::pair<size_t, float>> log_prob_idx =
-            get_pruned_log_probs(prob, cutoff_prob, cutoff_top_n);
-        // loop over chars
-        size_t log_prob_idx_len = log_prob_idx.size();
-        for (size_t index = 0; index < log_prob_idx_len; index++) {
-            SearchOneChar(full_beam, log_prob_idx[index], min_cutoff);
-        }
-        prefixes_.clear();
-        // update log probs
-        root_->iterate_to_vec(prefixes_);
-        // only preserve top beam_size prefixes_
-        if (prefixes_.size() >= beam_size) {
-            std::nth_element(prefixes_.begin(),
-                             prefixes_.begin() + beam_size,
-                             prefixes_.end(),
-                             prefix_compare);
-            for (size_t i = beam_size; i < prefixes_.size(); ++i) {
-                prefixes_[i]->remove();
-            }
-        }  // end if
-        num_frame_decoded_++;
-    }  // end for probs_seq
-}
-int32 CTCBeamSearch::SearchOneChar(
-    const bool& full_beam,
-    const std::pair<size_t, BaseFloat>& log_prob_idx,
-    const BaseFloat& min_cutoff) {
-    size_t beam_size = opts_.beam_size;
-    const auto& c = log_prob_idx.first;
-    const auto& log_prob_c = log_prob_idx.second;
-    size_t prefixes_len = std::min(prefixes_.size(), beam_size);
-    for (size_t i = 0; i < prefixes_len; ++i) {
-        auto prefix = prefixes_[i];
-        if (full_beam && log_prob_c + prefix->score < min_cutoff) {
-            break;
-        }
-        if (c == opts_.blank) {
-            prefix->log_prob_b_cur =
-                log_sum_exp(prefix->log_prob_b_cur, log_prob_c + prefix->score);
-            continue;
-        }
-        // repeated character
-        if (c == prefix->character) {
-            // p_{nb}(l;x_{1:t}) = p(c;x_{t})p(l;x_{1:t-1})
-            prefix->log_prob_nb_cur = log_sum_exp(
-                prefix->log_prob_nb_cur, log_prob_c + prefix->log_prob_nb_prev);
-        }
-        // get new prefix
-        auto prefix_new = prefix->get_path_trie(c);
-        if (prefix_new != nullptr) {
-            float log_p = -NUM_FLT_INF;
-            if (c == prefix->character &&
-                prefix->log_prob_b_prev > -NUM_FLT_INF) {
-                // p_{nb}(l^{+};x_{1:t}) = p(c;x_{t})p_{b}(l;x_{1:t-1})
-                log_p = log_prob_c + prefix->log_prob_b_prev;
-            } else if (c != prefix->character) {
-                // p_{nb}(l^{+};x_{1:t}) = p(c;x_{t}) p(l;x_{1:t-1})
-                log_p = log_prob_c + prefix->score;
-            }
-            // language model scoring
-            if (init_ext_scorer_ != nullptr &&
-                (c == space_id_ || init_ext_scorer_->is_character_based())) {
-                PathTrie* prefix_to_score = nullptr;
-                // skip scoring the space
-                if (init_ext_scorer_->is_character_based()) {
-                    prefix_to_score = prefix_new;
-                } else {
-                    prefix_to_score = prefix;
-                }
-                float score = 0.0;
-                vector<string> ngram;
-                ngram = init_ext_scorer_->make_ngram(prefix_to_score);
-                // lm score: p_{lm}(W)^{\alpha} + \beta
-                score = init_ext_scorer_->get_log_cond_prob(ngram) *
-                        init_ext_scorer_->alpha;
-                log_p += score;
-                log_p += init_ext_scorer_->beta;
-            }
-            // p_{nb}(l;x_{1:t})
-            prefix_new->log_prob_nb_cur =
-                log_sum_exp(prefix_new->log_prob_nb_cur, log_p);
-        }
-    }  // end of loop over prefix
-    return 0;
-}
-void CTCBeamSearch::CalculateApproxScore() {
-    size_t beam_size = opts_.beam_size;
-    size_t num_prefixes_ = std::min(prefixes_.size(), beam_size);
-    std::sort(
-        prefixes_.begin(), prefixes_.begin() + num_prefixes_, prefix_compare);
-    // compute aproximate ctc score as the return score, without affecting the
-    // return order of decoding result. To delete when decoder gets stable.
-    for (size_t i = 0; i < beam_size && i < prefixes_.size(); ++i) {
-        double approx_ctc = prefixes_[i]->score;
-        if (init_ext_scorer_ != nullptr) {
-            vector<int> output;
-            prefixes_[i]->get_path_vec(output);
-            auto prefix_length = output.size();
-            auto words = init_ext_scorer_->split_labels(output);
-            // remove word insert
-            approx_ctc = approx_ctc - prefix_length * init_ext_scorer_->beta;
-            // remove language model weight:
-            approx_ctc -= (init_ext_scorer_->get_sent_log_prob(words)) *
-                          init_ext_scorer_->alpha;
-        }
-        prefixes_[i]->approx_ctc = approx_ctc;
-    }
-}
-void CTCBeamSearch::LMRescore() {
-    size_t beam_size = opts_.beam_size;
-    if (init_ext_scorer_ != nullptr &&
-        !init_ext_scorer_->is_character_based()) {
-        for (size_t i = 0; i < beam_size && i < prefixes_.size(); ++i) {
-            auto prefix = prefixes_[i];
-            if (!prefix->is_empty() && prefix->character != space_id_) {
-                float score = 0.0;
-                vector<string> ngram = init_ext_scorer_->make_ngram(prefix);
-                score = init_ext_scorer_->get_log_cond_prob(ngram) *
-                        init_ext_scorer_->alpha;
-                score += init_ext_scorer_->beta;
-                prefix->score += score;
-            }
-        }
-    }
-}
-}  // namespace ppspeech
--- a/speechx/speechx/asr/decoder/ctc_beam_search_decoder.h
+++ b/speechx/speechx/asr/decoder/ctc_beam_search_decoder.h
-// Copyright (c) 2022 PaddlePaddle Authors. All Rights Reserved.
-//
-// Licensed under the Apache License, Version 2.0 (the "License");
-// you may not use this file except in compliance with the License.
-// You may obtain a copy of the License at
-//
-//     http://www.apache.org/licenses/LICENSE-2.0
-//
-// Unless required by applicable law or agreed to in writing, software
-// distributed under the License is distributed on an "AS IS" BASIS,
-// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
-// See the License for the specific language governing permissions and
-// limitations under the License.
-// used by deepspeech2
-#pragma once
-#include "decoder/ctc_beam_search_opt.h"
-#include "decoder/ctc_decoders/path_trie.h"
-#include "decoder/ctc_decoders/scorer.h"
-#include "decoder/decoder_itf.h"
-namespace ppspeech {
-class CTCBeamSearch : public DecoderBase {
-  public:
-    explicit CTCBeamSearch(const CTCBeamSearchOptions& opts);
-    ~CTCBeamSearch() {}
-    void InitDecoder();
-    void Reset();
-    void AdvanceDecode(
-        const std::shared_ptr<kaldi::DecodableInterface>& decodable);
-    void Decode(std::shared_ptr<kaldi::DecodableInterface> decodable);
-    std::string GetBestPath();
-    std::vector<std::pair<double, std::string>> GetNBestPath();
-    std::vector<std::pair<double, std::string>> GetNBestPath(int n);
-    std::string GetFinalBestPath();
-    std::string GetPartialResult() {
-        CHECK(false) << "Not implement.";
-        return {};
-    }
-    int DecodeLikelihoods(const std::vector<std::vector<BaseFloat>>& probs,
-                          const std::vector<std::string>& nbest_words);
-  private:
-    void ResetPrefixes();
-    int32 SearchOneChar(const bool& full_beam,
-                        const std::pair<size_t, BaseFloat>& log_prob_idx,
-                        const BaseFloat& min_cutoff);
-    void CalculateApproxScore();
-    void LMRescore();
-    void AdvanceDecoding(const std::vector<std::vector<BaseFloat>>& probs);
-    CTCBeamSearchOptions opts_;
-    std::shared_ptr<Scorer> init_ext_scorer_;  // todo separate later
-    std::vector<std::string> vocabulary_;      // todo remove later
-    int space_id_;
-    std::shared_ptr<PathTrie> root_;
-    std::vector<PathTrie*> prefixes_;
-    DISALLOW_COPY_AND_ASSIGN(CTCBeamSearch);
-};
-}  // namespace ppspeech
\ No newline at end of file
--- a/speechx/speechx/asr/decoder/ctc_beam_search_decoder_main.cc
+++ b/speechx/speechx/asr/decoder/ctc_beam_search_decoder_main.cc
-// Copyright (c) 2022 PaddlePaddle Authors. All Rights Reserved.
-//
-// Licensed under the Apache License, Version 2.0 (the "License");
-// you may not use this file except in compliance with the License.
-// You may obtain a copy of the License at
-//
-//     http://www.apache.org/licenses/LICENSE-2.0
-//
-// Unless required by applicable law or agreed to in writing, software
-// distributed under the License is distributed on an "AS IS" BASIS,
-// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
-// See the License for the specific language governing permissions and
-// limitations under the License.
-// used by deepspeech2
-#include "base/flags.h"
-#include "base/log.h"
-#include "decoder/ctc_beam_search_decoder.h"
-#include "frontend/audio/data_cache.h"
-#include "kaldi/util/table-types.h"
-#include "nnet/decodable.h"
-#include "nnet/ds2_nnet.h"
-DEFINE_string(feature_rspecifier, "", "test feature rspecifier");
-DEFINE_string(result_wspecifier, "", "test result wspecifier");
-DEFINE_string(model_path, "avg_1.jit.pdmodel", "paddle nnet model");
-DEFINE_string(param_path, "avg_1.jit.pdiparams", "paddle nnet model param");
-DEFINE_string(dict_file, "vocab.txt", "vocabulary of lm");
-DEFINE_string(lm_path, "", "language model");
-DEFINE_int32(receptive_field_length,
-             7,
-             "receptive field of two CNN(kernel=3) downsampling module.");
-DEFINE_int32(subsampling_rate,
-             4,
-             "two CNN(kernel=3) module downsampling rate.");
-DEFINE_string(
-    model_input_names,
-    "audio_chunk,audio_chunk_lens,chunk_state_h_box,chunk_state_c_box",
-    "model input names");
-DEFINE_string(model_output_names,
-              "softmax_0.tmp_0,tmp_5,concat_0.tmp_0,concat_1.tmp_0",
-              "model output names");
-DEFINE_string(model_cache_names,
-              "chunk_state_h_box,chunk_state_c_box",
-              "model cache names");
-DEFINE_string(model_cache_shapes, "5-1-1024,5-1-1024", "model cache shapes");
-DEFINE_int32(nnet_decoder_chunk, 1, "paddle nnet forward chunk");
-using kaldi::BaseFloat;
-using kaldi::Matrix;
-using std::vector;
-// test ds2 online decoder by feeding speech feature
-int main(int argc, char* argv[]) {
-    gflags::SetUsageMessage("Usage:");
-    gflags::ParseCommandLineFlags(&argc, &argv, false);
-    google::InitGoogleLogging(argv[0]);
-    google::InstallFailureSignalHandler();
-    FLAGS_logtostderr = 1;
-    CHECK_NE(FLAGS_result_wspecifier, "");
-    CHECK_NE(FLAGS_feature_rspecifier, "");
-    kaldi::SequentialBaseFloatMatrixReader feature_reader(
-        FLAGS_feature_rspecifier);
-    kaldi::TokenWriter result_writer(FLAGS_result_wspecifier);
-    std::string model_path = FLAGS_model_path;
-    std::string model_params = FLAGS_param_path;
-    std::string dict_file = FLAGS_dict_file;
-    std::string lm_path = FLAGS_lm_path;
-    LOG(INFO) << "model path: " << model_path;
-    LOG(INFO) << "model param: " << model_params;
-    LOG(INFO) << "dict path: " << dict_file;
-    LOG(INFO) << "lm path: " << lm_path;
-    int32 num_done = 0, num_err = 0;
-    ppspeech::CTCBeamSearchOptions opts;
-    opts.dict_file = dict_file;
-    opts.lm_path = lm_path;
-    ppspeech::CTCBeamSearch decoder(opts);
-    ppspeech::ModelOptions model_opts = ppspeech::ModelOptions::InitFromFlags();
-    std::shared_ptr<ppspeech::PaddleNnet> nnet(
-        new ppspeech::PaddleNnet(model_opts));
-    std::shared_ptr<ppspeech::DataCache> raw_data(new ppspeech::DataCache());
-    std::shared_ptr<ppspeech::Decodable> decodable(
-        new ppspeech::Decodable(nnet, raw_data));
-    int32 chunk_size = FLAGS_receptive_field_length +
-                       (FLAGS_nnet_decoder_chunk - 1) * FLAGS_subsampling_rate;
-    int32 chunk_stride = FLAGS_subsampling_rate * FLAGS_nnet_decoder_chunk;
-    int32 receptive_field_length = FLAGS_receptive_field_length;
-    LOG(INFO) << "chunk size (frame): " << chunk_size;
-    LOG(INFO) << "chunk stride (frame): " << chunk_stride;
-    LOG(INFO) << "receptive field (frame): " << receptive_field_length;
-    decoder.InitDecoder();
-    kaldi::Timer timer;
-    for (; !feature_reader.Done(); feature_reader.Next()) {
-        string utt = feature_reader.Key();
-        kaldi::Matrix<BaseFloat> feature = feature_reader.Value();
-        raw_data->SetDim(feature.NumCols());
-        LOG(INFO) << "process utt: " << utt;
-        LOG(INFO) << "rows: " << feature.NumRows();
-        LOG(INFO) << "cols: " << feature.NumCols();
-        int32 row_idx = 0;
-        int32 padding_len = 0;
-        int32 ori_feature_len = feature.NumRows();
-        if ((feature.NumRows() - chunk_size) % chunk_stride != 0) {
-            padding_len =
-                chunk_stride - (feature.NumRows() - chunk_size) % chunk_stride;
-            feature.Resize(feature.NumRows() + padding_len,
-                           feature.NumCols(),
-                           kaldi::kCopyData);
-        }
-        int32 num_chunks = (feature.NumRows() - chunk_size) / chunk_stride + 1;
-        for (int chunk_idx = 0; chunk_idx < num_chunks; ++chunk_idx) {
-            kaldi::Vector<kaldi::BaseFloat> feature_chunk(chunk_size *
-                                                          feature.NumCols());
-            int32 feature_chunk_size = 0;
-            if (ori_feature_len > chunk_idx * chunk_stride) {
-                feature_chunk_size = std::min(
-                    ori_feature_len - chunk_idx * chunk_stride, chunk_size);
-            }
-            if (feature_chunk_size < receptive_field_length) break;
-            int32 start = chunk_idx * chunk_stride;
-            for (int row_id = 0; row_id < chunk_size; ++row_id) {
-                kaldi::SubVector<kaldi::BaseFloat> tmp(feature, start);
-                kaldi::SubVector<kaldi::BaseFloat> f_chunk_tmp(
-                    feature_chunk.Data() + row_id * feature.NumCols(),
-                    feature.NumCols());
-                f_chunk_tmp.CopyFromVec(tmp);
-                ++start;
-            }
-            raw_data->Accept(feature_chunk);
-            if (chunk_idx == num_chunks - 1) {
-                raw_data->SetFinished();
-            }
-            decoder.AdvanceDecode(decodable);
-        }
-        std::string result;
-        result = decoder.GetFinalBestPath();
-        decodable->Reset();
-        decoder.Reset();
-        if (result.empty()) {
-            // the TokenWriter can not write empty string.
-            ++num_err;
-            KALDI_LOG << " the result of " << utt << " is empty";
-            continue;
-        }
-        KALDI_LOG << " the result of " << utt << " is " << result;
-        result_writer.Write(utt, result);
-        ++num_done;
-    }
-    KALDI_LOG << "Done " << num_done << " utterances, " << num_err
-              << " with errors.";
-    double elapsed = timer.Elapsed();
-    KALDI_LOG << " cost:" << elapsed << " s";
-    return (num_done != 0 ? 0 : 1);
-}
--- a/speechx/speechx/asr/decoder/ctc_decoders/.gitignore
+++ b/speechx/speechx/asr/decoder/ctc_decoders/.gitignore
-ThreadPool/
-build/
-dist/
-kenlm/
-openfst-1.6.3/
-openfst-1.6.3.tar.gz
-swig_decoders.egg-info/
-decoders_wrap.cxx
-swig_decoders.py
--- a/speechx/speechx/asr/decoder/ctc_decoders/ctc_beam_search_decoder.cpp
+++ b/speechx/speechx/asr/decoder/ctc_decoders/ctc_beam_search_decoder.cpp
--- a/speechx/speechx/asr/decoder/ctc_decoders/ctc_beam_search_decoder.h
+++ b/speechx/speechx/asr/decoder/ctc_decoders/ctc_beam_search_decoder.h
--- a/speechx/speechx/asr/decoder/ctc_decoders/ctc_greedy_decoder.cpp
+++ b/speechx/speechx/asr/decoder/ctc_decoders/ctc_greedy_decoder.cpp
-// Copyright (c) 2021 PaddlePaddle Authors. All Rights Reserved.
-//
-// Licensed under the Apache License, Version 2.0 (the "COPYING.APACHE2.0");
-// you may not use this file except in compliance with the License.
-// You may obtain a copy of the License at
-//
-//     http://www.apache.org/licenses/LICENSE-2.0
-//
-// Unless required by applicable law or agreed to in writing, software
-// distributed under the License is distributed on an "AS IS" BASIS,
-// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
-// See the License for the specific language governing permissions and
-// limitations under the License.
-#include "ctc_greedy_decoder.h"
-#include "decoder_utils.h"
-std::string ctc_greedy_decoding(
-    const std::vector<std::vector<double>> &probs_seq,
-    const std::vector<std::string> &vocabulary,
-    size_t blank_id) {
-    // dimension check
-    size_t num_time_steps = probs_seq.size();
-    for (size_t i = 0; i < num_time_steps; ++i) {
-        VALID_CHECK_EQ(probs_seq[i].size(),
-                       vocabulary.size(),
-                       "The shape of probs_seq does not match with "
-                       "the shape of the vocabulary");
-    }
-    // size_t blank_id = vocabulary.size();
-    std::vector<size_t> max_idx_vec(num_time_steps, 0);
-    std::vector<size_t> idx_vec;
-    for (size_t i = 0; i < num_time_steps; ++i) {
-        double max_prob = 0.0;
-        size_t max_idx = 0;
-        const std::vector<double> &probs_step = probs_seq[i];
-        for (size_t j = 0; j < probs_step.size(); ++j) {
-            if (max_prob < probs_step[j]) {
-                max_idx = j;
-                max_prob = probs_step[j];
-            }
-        }
-        // id with maximum probability in current time step
-        max_idx_vec[i] = max_idx;
-        // deduplicate
-        if ((i == 0) || ((i > 0) && max_idx_vec[i] != max_idx_vec[i - 1])) {
-            idx_vec.push_back(max_idx_vec[i]);
-        }
-    }
-    std::string best_path_result;
-    for (size_t i = 0; i < idx_vec.size(); ++i) {
-        if (idx_vec[i] != blank_id) {
-            std::string ch = vocabulary[idx_vec[i]];
-            best_path_result += (ch == kSPACE) ? tSPACE : ch;
-        }
-    }
-    return best_path_result;
-}
--- a/speechx/speechx/asr/decoder/ctc_decoders/ctc_greedy_decoder.h
+++ b/speechx/speechx/asr/decoder/ctc_decoders/ctc_greedy_decoder.h
-// Copyright (c) 2021 PaddlePaddle Authors. All Rights Reserved.
-//
-// Licensed under the Apache License, Version 2.0 (the "COPYING.APACHE2.0");
-// you may not use this file except in compliance with the License.
-// You may obtain a copy of the License at
-//
-//     http://www.apache.org/licenses/LICENSE-2.0
-//
-// Unless required by applicable law or agreed to in writing, software
-// distributed under the License is distributed on an "AS IS" BASIS,
-// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
-// See the License for the specific language governing permissions and
-// limitations under the License.
-#ifndef CTC_GREEDY_DECODER_H
-#define CTC_GREEDY_DECODER_H
-#include <string>
-#include <vector>
-/* CTC Greedy (Best Path) Decoder
- *
- * Parameters:
- *     probs_seq: 2-D vector that each element is a vector of probabilities
- *               over vocabulary of one time step.
- *     vocabulary: A vector of vocabulary.
- * Return:
- *     The decoding result in string
- */
-std::string ctc_greedy_decoding(
-    const std::vector<std::vector<double>>& probs_seq,
-    const std::vector<std::string>& vocabulary,
-    size_t blank_id);
-#endif  // CTC_GREEDY_DECODER_H
--- a/speechx/speechx/asr/decoder/ctc_decoders/decoder_utils.cpp
+++ b/speechx/speechx/asr/decoder/ctc_decoders/decoder_utils.cpp
--- a/speechx/speechx/asr/decoder/ctc_decoders/decoder_utils.h
+++ b/speechx/speechx/asr/decoder/ctc_decoders/decoder_utils.h
--- a/speechx/speechx/asr/decoder/ctc_decoders/path_trie.cpp
+++ b/speechx/speechx/asr/decoder/ctc_decoders/path_trie.cpp
--- a/speechx/speechx/asr/decoder/ctc_decoders/path_trie.h
+++ b/speechx/speechx/asr/decoder/ctc_decoders/path_trie.h
--- a/speechx/speechx/asr/decoder/ctc_decoders/scorer.cpp
+++ b/speechx/speechx/asr/decoder/ctc_decoders/scorer.cpp
--- a/speechx/speechx/asr/decoder/ctc_decoders/scorer.h
+++ b/speechx/speechx/asr/decoder/ctc_decoders/scorer.h
--- a/speechx/speechx/asr/decoder/nnet_logprob_decoder_main.cc
+++ b/speechx/speechx/asr/decoder/nnet_logprob_decoder_main.cc
--- a/speechx/speechx/asr/decoder/param.h
+++ b/speechx/speechx/asr/decoder/param.h
@@ -15,8 +15,7 @@
 #pragma once
 #include "base/common.h"
-#include "decoder/ctc_beam_search_decoder.h"
+//#include "decoder/ctc_tlg_decoder.h"
-#include "decoder/ctc_tlg_decoder.h"
 // feature
 DEFINE_bool(use_fbank, false, "False for fbank; or linear feature");

--- a/speechx/speechx/asr/nnet/CMakeLists.txt
+++ b/speechx/speechx/asr/nnet/CMakeLists.txt
--- a/speechx/speechx/asr/nnet/ds2_nnet.cc
+++ b/speechx/speechx/asr/nnet/ds2_nnet.cc
--- a/speechx/speechx/asr/nnet/ds2_nnet.h
+++ b/speechx/speechx/asr/nnet/ds2_nnet.h
--- a/speechx/speechx/asr/nnet/ds2_nnet_main.cc
+++ b/speechx/speechx/asr/nnet/ds2_nnet_main.cc
--- a/speechx/speechx/asr/nnet/nnet_producer.cc
+++ b/speechx/speechx/asr/nnet/nnet_producer.cc
--- a/speechx/speechx/asr/recognizer/CMakeLists.txt
+++ b/speechx/speechx/asr/recognizer/CMakeLists.txt
--- a/speechx/speechx/asr/recognizer/recognizer.cc
+++ b/speechx/speechx/asr/recognizer/recognizer.cc
--- a/speechx/speechx/asr/recognizer/recognizer.h
+++ b/speechx/speechx/asr/recognizer/recognizer.h
--- a/speechx/speechx/asr/recognizer/recognizer_main.cc
+++ b/speechx/speechx/asr/recognizer/recognizer_main.cc
--- a/speechx/speechx/codelab/CMakeLists.txt
+++ b/speechx/speechx/codelab/CMakeLists.txt
--- a/speechx/speechx/codelab/nnet/CMakeLists.txt
+++ b/speechx/speechx/codelab/nnet/CMakeLists.txt
--- a/speechx/speechx/codelab/nnet/ds2_model_test_main.cc
+++ b/speechx/speechx/codelab/nnet/ds2_model_test_main.cc
--- a/speechx/speechx/common/frontend/audio/cmvn_json2kaldi_main.cc
+++ b/speechx/speechx/common/frontend/audio/cmvn_json2kaldi_main.cc
--- a/speechx/speechx/common/utils/picojson.h
+++ b/speechx/speechx/common/utils/picojson.h