#!/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