From dbebee86fdf65d69c88c92081ed4043c5b8f2f37 Mon Sep 17 00:00:00 2001
From: Binbin Zhang <binbzha@qq.com>
Date: Wed, 10 Nov 2021 18:57:52 +0800
Subject: [PATCH] [examples] support hi xiaowen dataset

---
 examples/hi_xiaowen/s0/README.md              |  10 ++
 examples/hi_xiaowen/s0/conf/ds_tcn.yaml       |  44 ++++++
 examples/hi_xiaowen/s0/conf/gru.yaml          |  41 ++++++
 examples/hi_xiaowen/s0/conf/tcn.yaml          |  44 ++++++
 examples/hi_xiaowen/s0/kws                    |   1 +
 .../s0/local/mobvoi_data_download.sh          |  65 +++++++++
 examples/hi_xiaowen/s0/local/prepare_data.py  |  43 ++++++
 examples/hi_xiaowen/s0/path.sh                |   5 +
 examples/hi_xiaowen/s0/run.sh                 | 126 ++++++++++++++++++
 examples/hi_xiaowen/s0/tools                  |   1 +
 10 files changed, 380 insertions(+)
 create mode 100644 examples/hi_xiaowen/s0/README.md
 create mode 100644 examples/hi_xiaowen/s0/conf/ds_tcn.yaml
 create mode 100644 examples/hi_xiaowen/s0/conf/gru.yaml
 create mode 100644 examples/hi_xiaowen/s0/conf/tcn.yaml
 create mode 120000 examples/hi_xiaowen/s0/kws
 create mode 100755 examples/hi_xiaowen/s0/local/mobvoi_data_download.sh
 create mode 100755 examples/hi_xiaowen/s0/local/prepare_data.py
 create mode 100755 examples/hi_xiaowen/s0/path.sh
 create mode 100755 examples/hi_xiaowen/s0/run.sh
 create mode 120000 examples/hi_xiaowen/s0/tools

diff --git a/examples/hi_xiaowen/s0/README.md b/examples/hi_xiaowen/s0/README.md
new file mode 100644
index 0000000..15084d5
--- /dev/null
+++ b/examples/hi_xiaowen/s0/README.md
@@ -0,0 +1,10 @@
+FRRs with FAR fixed at once per hour:
+
+| model            | params(K) | epoch     | hi_xiaowen | nihao_wenwen |
+|------------------|-----------|-----------|------------|--------------|
+| GRU              | 203       | 80(avg30) | 0.088901   | 0.083827     |
+| TCN              | 134       | 80(avg30) | 0.023494   | 0.029884     |
+| DS_TCN           | 21        | 60        | 0.011559   | 0.014190     |
+| DS_TCN           | 21        | 80        | 0.010807   | 0.014754     |
+| DS_TCN           | 21        | 80(avg30) | 0.009867   | 0.014472     |
+| DS_TCN(spec_aug) | 21        | 80(avg30) | 0.029039   | 0.022648     |
diff --git a/examples/hi_xiaowen/s0/conf/ds_tcn.yaml b/examples/hi_xiaowen/s0/conf/ds_tcn.yaml
new file mode 100644
index 0000000..7da0181
--- /dev/null
+++ b/examples/hi_xiaowen/s0/conf/ds_tcn.yaml
@@ -0,0 +1,44 @@
+dataset_conf:
+    filter_conf:
+        max_length: 2048
+        min_length: 0
+    resample_conf:
+        resample_rate: 16000
+    speed_perturb: false
+    fbank_conf:
+        num_mel_bins: 40
+        frame_shift: 10
+        frame_length: 25
+        dither: 0.1
+    spec_aug: true
+    spec_aug_conf:
+        num_t_mask: 1
+        num_f_mask: 1
+        max_t: 50
+        max_f: 30
+    shuffle: true
+    shuffle_conf:
+        shuffle_size: 1500
+    batch_conf:
+        batch_size: 256
+
+model:
+    hidden_dim: 64
+    subsampling:
+        type: linear
+    body:
+        type: tcn
+        ds: true
+        num_layers: 4
+        kernel_size: 8
+        dropout: 0.1
+
+optim: adam
+optim_conf:
+    lr: 0.001
+
+training_config:
+    grad_clip: 5
+    max_epoch: 80
+    log_interval: 10
+
diff --git a/examples/hi_xiaowen/s0/conf/gru.yaml b/examples/hi_xiaowen/s0/conf/gru.yaml
new file mode 100644
index 0000000..e664319
--- /dev/null
+++ b/examples/hi_xiaowen/s0/conf/gru.yaml
@@ -0,0 +1,41 @@
+dataset_conf:
+    filter_conf:
+        max_length: 2048
+        min_length: 0
+    resample_conf:
+        resample_rate: 16000
+    speed_perturb: false
+    fbank_conf:
+        num_mel_bins: 40
+        frame_shift: 10
+        frame_length: 25
+        dither: 0.1
+    spec_aug: false
+    spec_aug_conf:
+        num_t_mask: 2
+        num_f_mask: 2
+        max_t: 50
+        max_f: 30
+    shuffle: true
+    shuffle_conf:
+        shuffle_size: 1500
+    batch_conf:
+        batch_size: 256
+
+model:
+    hidden_dim: 128
+    subsampling:
+        type: linear
+    body:
+        type: gru
+        num_layers: 2
+
+optim: adam
+optim_conf:
+    lr: 0.001
+
+training_config:
+    grad_clip: 5
+    max_epoch: 80
+    log_interval: 10
+
diff --git a/examples/hi_xiaowen/s0/conf/tcn.yaml b/examples/hi_xiaowen/s0/conf/tcn.yaml
new file mode 100644
index 0000000..0612634
--- /dev/null
+++ b/examples/hi_xiaowen/s0/conf/tcn.yaml
@@ -0,0 +1,44 @@
+dataset_conf:
+    filter_conf:
+        max_length: 2048
+        min_length: 0
+    resample_conf:
+        resample_rate: 16000
+    speed_perturb: false
+    fbank_conf:
+        num_mel_bins: 40
+        frame_shift: 10
+        frame_length: 25
+        dither: 0.1
+    spec_aug: false
+    spec_aug_conf:
+        num_t_mask: 2
+        num_f_mask: 2
+        max_t: 50
+        max_f: 30
+    shuffle: true
+    shuffle_conf:
+        shuffle_size: 1500
+    batch_conf:
+        batch_size: 256
+
+model:
+    hidden_dim: 64
+    subsampling:
+        type: linear
+    body:
+        type: tcn
+        ds: false
+        num_layers: 4
+        kernel_size: 8
+        dropout: 0.1
+
+optim: adam
+optim_conf:
+    lr: 0.001
+
+training_config:
+    grad_clip: 5
+    max_epoch: 80
+    log_interval: 10
+
diff --git a/examples/hi_xiaowen/s0/kws b/examples/hi_xiaowen/s0/kws
new file mode 120000
index 0000000..7a3e8e1
--- /dev/null
+++ b/examples/hi_xiaowen/s0/kws
@@ -0,0 +1 @@
+../../../kws
\ No newline at end of file
diff --git a/examples/hi_xiaowen/s0/local/mobvoi_data_download.sh b/examples/hi_xiaowen/s0/local/mobvoi_data_download.sh
new file mode 100755
index 0000000..7ba3709
--- /dev/null
+++ b/examples/hi_xiaowen/s0/local/mobvoi_data_download.sh
@@ -0,0 +1,65 @@
+#!/bin/bash
+
+# Copyright  2018-2020  Yiming Wang
+#            2018-2020  Daniel Povey
+#            2021       Binbin Zhang
+
+[ -f ./path.sh ] && . ./path.sh
+
+dl_dir=data/download
+
+. tools/parse_options.sh || exit 1;
+
+mkdir -p $dl_dir
+
+dataset=mobvoi_hotword_dataset.tgz
+resources=mobvoi_hotword_dataset_resources.tgz
+
+# base url for downloads.
+data_url=http://www.openslr.org/resources/87
+
+if [[ $(hostname -f) == *.clsp.jhu.edu ]]; then
+  src_path=/export/fs04/a07/ywang/mobvoihotwords
+else
+  src_path=$dl_dir
+fi
+
+if [ ! -f $src_path/$dataset ] || [ ! -f $src_path/$resources ]; then
+  if ! which wget >/dev/null; then
+    echo "$0: wget is not installed."
+    exit 1;
+  fi
+
+  if [ ! -f $src_path/$dataset ]; then
+    echo "$0: downloading data from $data_url/$dataset.  This may take some time, please be patient."
+    if ! wget --no-check-certificate -O $dl_dir/$dataset $data_url/$dataset; then
+      echo "$0: error executing wget $data_url/$dataset"
+      exit 1;
+    fi
+  fi
+
+  if [ ! -f $src_path/$resources ]; then
+    if ! wget --no-check-certificate -O $dl_dir/$resources $data_url/$resources; then
+      echo "$0: error executing wget $data_url/$resources"
+      exit 1;
+    fi
+  fi
+fi
+
+if [ -d $dl_dir/$(basename "$dataset" .tgz) ]; then
+  echo "Not extracting $(basename "$dataset" .tgz) as it is already there."
+else
+  echo "Extracting $dataset..."
+  tar -xvzf $src_path/$dataset -C $dl_dir || exit 1;
+  echo "Done extracting $dataset."
+fi
+
+if [ -d $dl_dir/$(basename "$resources" .tgz) ]; then
+  echo "Not extracting $(basename "$dataset" .tar.gz) as it is already there."
+else
+  echo "Extracting $resources..."
+  tar -xvzf $src_path/$resources -C $dl_dir || exit 1;
+  echo "Done extracting $resources."
+fi
+
+exit 0
diff --git a/examples/hi_xiaowen/s0/local/prepare_data.py b/examples/hi_xiaowen/s0/local/prepare_data.py
new file mode 100755
index 0000000..ad75a7f
--- /dev/null
+++ b/examples/hi_xiaowen/s0/local/prepare_data.py
@@ -0,0 +1,43 @@
+#!/usr/bin/env python3
+
+# Copyright 2018-2020  Yiming Wang
+#           2018-2020  Daniel Povey
+#           2021       Binbin Zhang
+# Apache 2.0
+""" This script prepares the Mobvoi data into kaldi format.
+"""
+
+import argparse
+import os
+import json
+
+
+def main():
+    parser = argparse.ArgumentParser(description="""Prepare data.""")
+    parser.add_argument('wav_dir',
+                        type=str,
+                        help='dir containing all the wav files')
+    parser.add_argument('path', type=str, help='path to the json file')
+    parser.add_argument('out_dir', type=str, help='out dir')
+    args = parser.parse_args()
+
+    with open(args.path, 'r', encoding='utf-8') as f:
+        data = json.load(f)
+        utt_id, label = [], []
+        for entry in data:
+            utt_id.append(entry['utt_id'])
+            label.append(int(entry['keyword_id']))
+
+    abs_dir = os.path.abspath(args.wav_dir)
+    wav_path = os.path.join(args.out_dir, 'wav.scp')
+    text_path = os.path.join(args.out_dir, 'text')
+    with open(wav_path, 'w', encoding='utf-8') as f_wav, \
+         open(text_path, 'w', encoding='utf-8') as f_text:
+        for utt, l in zip(utt_id, label):
+            f_wav.write('{} {}\n'.format(utt,
+                                         os.path.join(abs_dir, utt + ".wav")))
+            f_text.write('{} {}\n'.format(utt, l))
+
+
+if __name__ == "__main__":
+    main()
diff --git a/examples/hi_xiaowen/s0/path.sh b/examples/hi_xiaowen/s0/path.sh
new file mode 100755
index 0000000..cf09584
--- /dev/null
+++ b/examples/hi_xiaowen/s0/path.sh
@@ -0,0 +1,5 @@
+export PATH=$PWD:$PATH
+
+# NOTE(kan-bayashi): Use UTF-8 in Python to avoid UnicodeDecodeError when LC_ALL=C
+export PYTHONIOENCODING=UTF-8
+export PYTHONPATH=../../:$PYTHONPATH
diff --git a/examples/hi_xiaowen/s0/run.sh b/examples/hi_xiaowen/s0/run.sh
new file mode 100755
index 0000000..650532f
--- /dev/null
+++ b/examples/hi_xiaowen/s0/run.sh
@@ -0,0 +1,126 @@
+#!/bin/bash
+# Copyright 2021  Binbin Zhang
+
+. ./path.sh
+
+export CUDA_VISIBLE_DEVICES="0,1,2,3,4,5,6,7"
+
+stage=0
+stop_stage=4
+num_keywords=2
+
+config=conf/ds_tcn.yaml
+norm_mean=true
+norm_var=true
+gpu_id=0
+
+checkpoint=
+dir=exp/ds_tcn
+
+num_average=30
+score_checkpoint=$dir/avg_${num_average}.pt
+
+download_dir=/export/expts6/binbinzhang/data/
+
+. tools/parse_options.sh || exit 1;
+
+set -euo pipefail
+
+if [ ${stage} -le -1 ] && [ ${stop_stage} -ge -1 ]; then
+  echo "Download and extracte all datasets"
+  local/mobvoi_data_download.sh --dl_dir $download_dir
+fi
+
+
+if [ ${stage} -le 0 ] && [ ${stop_stage} -ge 0 ]; then
+  echo "Preparing datasets..."
+  mkdir dict
+  echo "<filler> -1" > dict/words.txt
+  echo "Hi_Xiaowen 0" >> dict/words.txt
+  echo "Nihao_Wenwen 1" >> dict/words.txt
+
+  for folder in train dev eval; do
+    mkdir -p data/$folder
+    for prefix in p n; do
+      mkdir -p data/${prefix}_$folder
+      json_path=$download_dir/mobvoi_hotword_dataset_resources/${prefix}_$folder.json
+      if [ $folder = "eval" ]; then
+        json_path=$download_dir/mobvoi_hotword_dataset_resources/${prefix}_test.json
+      fi
+      local/prepare_data.py $download_dir/mobvoi_hotword_dataset $json_path \
+        data/${prefix}_$folder
+    done
+    cat data/p_$folder/wav.scp data/n_$folder/wav.scp > data/$folder/wav.scp
+    cat data/p_$folder/text data/n_$folder/text > data/$folder/text
+    rm -rf data/p_$folder data/n_$folder
+  done
+fi
+
+
+if [ ${stage} -le 1 ] && [ ${stop_stage} -ge 1 ]; then
+  echo "Compute CMVN and Format datasets"
+  tools/compute_cmvn_stats.py --num_workers 16 --train_config $config \
+    --in_scp data/train/wav.scp \
+    --out_cmvn data/train/global_cmvn
+
+  for x in train dev eval; do
+    tools/wav_to_duration.sh --nj 8 data/$x/wav.scp data/$x/wav.dur
+    tools/make_list.py data/$x/wav.scp data/$x/text \
+      data/$x/wav.dur data/$x/data.list
+  done
+fi
+
+
+if [ ${stage} -le 2 ] && [ ${stop_stage} -ge 2 ]; then
+  echo "Start training ..."
+  mkdir -p $dir
+  cmvn_opts=
+  $norm_mean && cmvn_opts="--cmvn_file data/train/global_cmvn"
+  $norm_var && cmvn_opts="$cmvn_opts --norm_var"
+  python kws/bin/train.py --gpu $gpu_id \
+    --config $config \
+    --train_data data/train/data.list \
+    --cv_data data/dev/data.list \
+    --model_dir $dir \
+    --num_workers 8 \
+    --num_keywords $num_keywords \
+    --min_duration 50 \
+    $cmvn_opts \
+    ${checkpoint:+--checkpoint $checkpoint}
+fi
+
+if [ ${stage} -le 3 ] && [ ${stop_stage} -ge 3 ]; then
+  # Do model average
+  python kws/bin/average_model.py \
+    --dst_model $score_checkpoint \
+    --src_path $dir  \
+    --num ${num_average} \
+    --val_best
+
+  # Compute posterior score
+  result_dir=$dir/test_$(basename $score_checkpoint)
+  mkdir -p $result_dir
+  python kws/bin/score.py --gpu -1 \
+    --config $dir/config.yaml \
+    --test_data data/eval/data.list \
+    --batch_size 256 \
+    --checkpoint $score_checkpoint \
+    --score_file $result_dir/score.txt
+
+  # Compute detection error tradeoff
+  for keyword in 0 1; do
+    python kws/bin/compute_det.py \
+      --keyword $keyword \
+      --test_data data/eval/data.list \
+      --score_file $result_dir/score.txt \
+      --stats_file $result_dir/stats.${keyword}.txt
+  done
+fi
+
+
+if [ ${stage} -le 4 ] && [ ${stop_stage} -ge 4 ]; then
+  python kws/bin/export_jit.py --config $dir/config.yaml \
+    --checkpoint $score_checkpoint \
+    --output_file $dir/final.zip \
+    --output_quant_file $dir/final.quant.zip
+fi
diff --git a/examples/hi_xiaowen/s0/tools b/examples/hi_xiaowen/s0/tools
new file mode 120000
index 0000000..c92f417
--- /dev/null
+++ b/examples/hi_xiaowen/s0/tools
@@ -0,0 +1 @@
+../../../tools
\ No newline at end of file