refactor aishell scripts

deepspeech/exps/deepspeech2/config.py

@@ -85,4 +85,6 @@ def get_cfg_defaults():
     """Get a yacs CfgNode object with default values for my_project."""
     # Return a clone so that the defaults will not be altered
     # This is for the "local variable" use pattern
-    return _C.clone()
+    config = _C.clone()
+    config.set_new_allowed(True)
+    return config

deepspeech/exps/u2/config.py

@@ -18,7 +18,6 @@ from deepspeech.exps.u2.model import U2Trainer
 from deepspeech.io.dataset import ManifestDataset
 from deepspeech.models.u2 import U2Model
 
-
 _C = CfgNode()
 
 _C.data = ManifestDataset.params()
@@ -34,4 +33,6 @@ def get_cfg_defaults():
     """Get a yacs CfgNode object with default values for my_project."""
     # Return a clone so that the defaults will not be altered
     # This is for the "local variable" use pattern
-    return _C.clone()
+    config = _C.clone()
+    config.set_new_allowed(True)
+    return config

deepspeech/frontend/normalizer.py

@@ -149,7 +149,7 @@ class FeatureNormalizer(object):
                           eps=1e-20):
         """Compute mean and std from randomly sampled instances."""
         paddle.set_device('cpu')
-        
+
         collate_func = CollateFunc(featurize_func)
         dataset = AudioDataset(manifest_path, num_samples, self._rng)
         data_loader = DataLoader(

deepspeech/models/u2.py

@@ -62,7 +62,6 @@ class U2BaseModel(nn.Module):
         # network architecture
         default = CfgNode()
         # allow add new item when merge_with_file
-        default.set_new_allowed(True)
         default.cmvn_file = ""
         default.cmvn_file_type = "npz"
         default.input_dim = 0
@@ -85,7 +84,7 @@ class U2BaseModel(nn.Module):
                 # activation_type='swish',
                 # pos_enc_layer_type='rel_pos',
                 # selfattention_layer_type='rel_selfattn', 
-                ))
+            ))
         # decoder related
         default.decoder = 'transformer'
         default.decoder_conf = CfgNode(

examples/aishell/s1/conf/conformer.yaml

@@ -9,7 +9,7 @@ data:
   augmentation_config: conf/augmentation.json
   batch_size: 64
   min_input_len: 0.5
-  max_input_len: 20.0
+  max_input_len: 20.0 # second
   min_output_len: 0.0
   max_output_len: 400.0
   min_output_input_ratio: 0.05
@@ -30,7 +30,7 @@ data:
   keep_transcription_text: False
   sortagrad: True 
   shuffle_method: batch_shuffle
-  num_workers: 0
+  num_workers: 2
 
 
 # network architecture
@@ -75,7 +75,7 @@ model:
 
 training:
   n_epoch: 240
-  accum_grad: 1
+  accum_grad: 2
   global_grad_clip: 5.0
   optim: adam
   optim_conf:
@@ -89,7 +89,7 @@ training:
 
 
 decoding:
-  batch_size: 16
+  batch_size: 128
   error_rate_type: cer 
   decoding_method: attention  # 'attention', 'ctc_greedy_search', 'ctc_prefix_beam_search', 'attention_rescoring'
   lang_model_path: data/lm/common_crawl_00.prune01111.trie.klm

examples/aishell/s1/local/avg.sh

@@ -1,24 +1,23 @@
 #! /usr/bin/env bash
 
-if [ $# != 2 ];then
+if [ $# != 2 ]; then
     echo "usage: ${0} ckpt_dir avg_num"
     exit -1
 fi
 
-ckpt_path=${1}
+ckpt_dir=${1}
 average_num=${2}
-decode_checkpoint=${ckpt_path}/avg_${average_num}.pdparams
+decode_checkpoint=${ckpt_dir}/avg_${average_num}.pdparams
 
 python3 -u ${MAIN_ROOT}/utils/avg_model.py \
 --dst_model ${decode_checkpoint} \
---ckpt_dir ${ckpt_path}  \
+--ckpt_dir ${ckpt_dir}  \
 --num ${average_num} \
 --val_best
-            
+
 if [ $? -ne 0 ]; then
     echo "Failed in avg ckpt!"
     exit 1
 fi
 
-
 exit 0

examples/aishell/s1/local/export.sh

@@ -1,14 +1,28 @@
 #! /usr/bin/env bash
 
-if [ $# != 2 ];then
-    echo "usage: ${0}  ckpt_path jit_model_path"
+if [ $# != 3 ];then
+    echo "usage: $0 config_path ckpt_prefix jit_model_path"
     exit -1
 fi
 
+ngpu=$(echo $CUDA_VISIBLE_DEVICES | awk -F "," '{print NF}')
+echo "using $ngpu gpus..."
+
+config_path=$1
+ckpt_path_prefix=$2
+jit_model_export_path=$3
+
+device=gpu
+if [ ngpu == 0 ];then
+    device=cpu
+fi
+
 python3 -u ${BIN_DIR}/export.py \
---config conf/conformer.yaml \
---checkpoint_path ${1} \
---export_path ${2} 
+--device ${device} \
+--nproc ${ngpu} \
+--config ${config_path} \
+--checkpoint_path ${ckpt_path_prefix} \
+--export_path ${jit_model_export_path}
 
 
 if [ $? -ne 0 ]; then

examples/aishell/s1/local/test.sh

@@ -1,10 +1,20 @@
 #! /usr/bin/env bash
 
-if [ $# != 1 ];then
-    echo "usage: ${0} ckpt_path_prefix"
+if [ $# != 2 ];then
+    echo "usage: ${0} config_path ckpt_path_prefix"
     exit -1
 fi
 
+ngpu=$(echo $CUDA_VISIBLE_DEVICES | awk -F "," '{print NF}')
+echo "using $ngpu gpus..."
+
+device=gpu
+if [ ngpu == 0 ];then
+    device=cpu
+fi
+config_path=$1
+ckpt_prefix=$2
+
 # download language model
 #bash local/download_lm_ch.sh
 #if [ $? -ne 0 ]; then
@@ -12,11 +22,11 @@ fi
 #fi
 
 python3 -u ${BIN_DIR}/test.py \
---device 'gpu' \
+--device ${device} \
 --nproc 1 \
---config conf/conformer.yaml \
---result_file ${1}.rsl \
---checkpoint_path ${1} 
+--config ${config_path} \
+--result_file ${ckpt_prefix}.rsl \
+--checkpoint_path ${ckpt_prefix}
 
 if [ $? -ne 0 ]; then
     echo "Failed in evaluation!"

examples/aishell/s1/local/train.sh

@@ -1,25 +1,33 @@
 #! /usr/bin/env bash
 
-if [ $# != 1 ];then
-    echo "usage: ${0} ckpt_tag"
+if [ $# != 2 ];then
+    echo "usage: CUDA_VISIBLE_DEVICES=0 ${0} config_path ckpt_name"
     exit -1
 fi
 
-mkdir -p exp
-
 ngpu=$(echo $CUDA_VISIBLE_DEVICES | awk -F "," '{print NF}')
 echo "using $ngpu gpus..."
 
+config_path=$1
+ckpt_name=$2
+
+device=gpu
+if [ ngpu == 0 ];then
+    device=cpu
+fi
+echo "using ${device}..."
+
+mkdir -p exp
+
 python3 -u ${BIN_DIR}/train.py \
---device 'gpu' \
+--device ${device} \
 --nproc ${ngpu} \
---config conf/conformer.yaml \
---output exp/${1}
+--config ${config_path} \
+--output exp/${ckpt_name}
 
 if [ $? -ne 0 ]; then
     echo "Failed in training!"
     exit 1
 fi
 
-
 exit 0

examples/aishell/s1/run.sh

@@ -1,19 +1,37 @@
 #!/bin/bash
-
+set -e
 source path.sh
-# only demos
 
-# prepare data
-bash ./local/data.sh
+stage=0
+stop_stage=100
+conf_path=conf/conformer.yaml
+ckpt=$(basename ${conf_path} | awk -F'.' '{print $1}')
+avg_num=20
+avg_ckpt=avg_${avg_num}
+
+source ${MAIN_ROOT}/utils/parse_options.sh || exit 1;
+
+if [ ${stage} -le 0 ] && [ ${stop_stage} -ge 0 ]; then
+    # prepare data
+    bash ./local/data.sh || exit -1
+fi
 
-# train model
-CUDA_VISIBLE_DEVICES=0,1,2,3 bash ./local/train.sh
+if [ ${stage} -le 1 ] && [ ${stop_stage} -ge 1 ]; then
+    # train model, all `ckpt` under `exp` dir
+    CUDA_VISIBLE_DEVICES=0,1,2,3 ./local/train.sh ${conf_path}  ${ckpt}
+fi
 
-# test model
-CUDA_VISIBLE_DEVICES=0 bash ./local/test.sh
+if [ ${stage} -le 2 ] && [ ${stop_stage} -ge 2 ]; then
+    # avg n best model
+    ./local/avg.sh exp/${ckpt}/checkpoints ${avg_num}
+fi
 
-# infer model
-CUDA_VISIBLE_DEVICES=0 bash ./local/infer.sh ckpt/checkpoints/step-3284
+if [ ${stage} -le 3 ] && [ ${stop_stage} -ge 3 ]; then
+    # test ckpt avg_n
+    CUDA_VISIBLE_DEVICES=4 ./local/test.sh ${conf_path} exp/${ckpt}/checkpoints/${avg_ckpt} || exit -1
+fi
 
-# export model
-bash ./local/export.sh ckpt/checkpoints/step-3284 jit.model
+if [ ${stage} -le 4 ] && [ ${stop_stage} -ge 4 ]; then
+    # export ckpt avg_n
+    CUDA_VISIBLE_DEVICES= ./local/export.sh ${conf_path} exp/${ckpt}/checkpoints/${avg_ckpt} exp/${ckpt}/checkpoints/${avg_ckpt}.jit
+fi