add librispeech scripts

examples/librispeech/s1/conf/chunk_confermer.yaml

@@ -79,7 +79,7 @@ model:
 
 
 training:
-  n_epoch: 20
+  n_epoch: 120
   accum_grad: 1
   global_grad_clip: 5.0
   optim: adam
@@ -90,11 +90,11 @@ training:
   scheduler_conf:
     warmup_steps: 25000
     lr_decay: 1.0
-  log_interval: 1
+  log_interval: 100
 
 
 decoding:
-  batch_size: 64
+  batch_size: 128
   error_rate_type: wer
   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/librispeech/s1/conf/chunk_transformer.yaml

@@ -8,7 +8,7 @@ data:
   spm_model_prefix: 'data/bpe_unigram_200'
   mean_std_filepath: ""
   augmentation_config: conf/augmentation.json
-  batch_size: 4
+  batch_size: 64
   min_input_len: 0.5  # second
   max_input_len: 20.0 # second
   min_output_len: 0.0 # tokens
@@ -72,18 +72,18 @@ model:
 
 
 training:
-  n_epoch: 20
+  n_epoch: 120
   accum_grad: 1
   global_grad_clip: 5.0
   optim: adam
   optim_conf:
-    lr: 0.002
+    lr: 0.001
     weight_decay: 1e-06
   scheduler: warmuplr     # pytorch v1.1.0+ required
   scheduler_conf:
     warmup_steps: 25000
     lr_decay: 1.0
-  log_interval: 1
+  log_interval: 100
 
 
 decoding:

examples/librispeech/s1/conf/conformer.yaml

@@ -5,14 +5,14 @@ data:
   test_manifest: data/manifest.tiny
   vocab_filepath: data/vocab.txt 
   unit_type: 'spm'
-  spm_model_prefix: 'data/bpe_unigram_200'
+  spm_model_prefix: 'data/bpe_unigram_5000'
   mean_std_filepath: ""
   augmentation_config: conf/augmentation.json
-  batch_size: 4
-  min_input_len: 0.5
-  max_input_len: 20.0
-  min_output_len: 0.0
-  max_output_len: 400.0
+  batch_size: 64
+  min_input_len: 0.5  # seconds
+  max_input_len: 20.0 # seconds
+  min_output_len: 0.0 # tokens
+  max_output_len: 400.0 # tokens
   min_output_input_ratio: 0.05
   max_output_input_ratio: 10.0
   raw_wav: True  # use raw_wav or kaldi feature
@@ -49,7 +49,7 @@ model:
         positional_dropout_rate: 0.1
         attention_dropout_rate: 0.0
         input_layer: conv2d # encoder input type, you can chose conv2d, conv2d6 and conv2d8
-        normalize_before: true
+        normalize_before: True
         use_cnn_module: True
         cnn_module_kernel: 15
         activation_type: 'swish'
@@ -75,18 +75,18 @@ model:
 
 
 training:
-  n_epoch: 20
-  accum_grad: 4
+  n_epoch: 120
+  accum_grad: 2
   global_grad_clip: 5.0
   optim: adam
   optim_conf:
-    lr: 0.002
+    lr: 0.004
     weight_decay: 1e-06
   scheduler: warmuplr     # pytorch v1.1.0+ required
   scheduler_conf:
     warmup_steps: 25000
     lr_decay: 1.0
-  log_interval: 1
+  log_interval: 100
 
 
 decoding:

examples/librispeech/s1/conf/transformer.yaml

@@ -8,7 +8,7 @@ data:
   spm_model_prefix: 'data/bpe_unigram_200'
   mean_std_filepath: ""
   augmentation_config: conf/augmentation.json
-  batch_size: 4
+  batch_size: 64
   min_input_len: 0.5  # second
   max_input_len: 20.0 # second
   min_output_len: 0.0 # tokens
@@ -70,18 +70,18 @@ model:
 
 
 training:
-  n_epoch: 20
-  accum_grad: 1
+  n_epoch: 120
+  accum_grad: 2
   global_grad_clip: 5.0
   optim: adam
   optim_conf:
-    lr: 0.002
+    lr: 0.004
     weight_decay: 1e-06
   scheduler: warmuplr     # pytorch v1.1.0+ required
   scheduler_conf:
     warmup_steps: 25000
     lr_decay: 1.0
-  log_interval: 1
+  log_interval: 100
 
 
 decoding:

examples/librispeech/s1/local/data.sh

@@ -4,7 +4,7 @@ stage=-1
 stop_stage=100
 
 # bpemode (unigram or bpe)
-nbpe=200
+nbpe=5000
 bpemode=unigram
 bpeprefix="data/bpe_${bpemode}_${nbpe}"
 
@@ -20,14 +20,16 @@ if [ ${stage} -le -1 ] && [ ${stop_stage} -ge -1 ]; then
     python3 ${TARGET_DIR}/librispeech/librispeech.py \
     --manifest_prefix="data/manifest" \
     --target_dir="${TARGET_DIR}/librispeech" \
-    --full_download="False"
+    --full_download="True"
 
     if [ $? -ne 0 ]; then
         echo "Prepare LibriSpeech failed. Terminated."
         exit 1
     fi
 
-    head -n 64 data/manifest.dev-clean  > data/manifest.tiny.raw
+    for set in train-clean-100 train-clean-360 train-other-500; do
+        cat data/manifest.${set} >> data/manifest.train.raw
+    done
 fi
 
 if [ ${stage} -le 0 ] && [ ${stop_stage} -ge 0 ]; then
@@ -38,7 +40,7 @@ if [ ${stage} -le 0 ] && [ ${stop_stage} -ge 0 ]; then
     --spm_mode ${bpemode} \
     --spm_model_prefix ${bpeprefix} \
     --vocab_path="data/vocab.txt" \
-    --manifest_paths="data/manifest.tiny.raw"
+    --manifest_paths="data/manifest.train.raw"
 
     if [ $? -ne 0 ]; then
         echo "Build vocabulary failed. Terminated."
@@ -49,16 +51,17 @@ fi
 
 if [ ${stage} -le 1 ] && [ ${stop_stage} -ge 1 ]; then
     # compute mean and stddev for normalizer
+    num_workers=$(nproc)
     python3 ${MAIN_ROOT}/utils/compute_mean_std.py \
-    --manifest_path="data/manifest.tiny.raw" \
-    --num_samples=64 \
+    --manifest_path="data/manifest.train.raw" \
+    --num_samples=-1 \
     --specgram_type="fbank" \
     --feat_dim=80 \
     --delta_delta=false \
     --sample_rate=16000 \
     --stride_ms=10.0 \
     --window_ms=25.0 \
-    --num_workers=2 \
+    --num_workers=${num_workers} \
     --output_path="data/mean_std.json"
 
     if [ $? -ne 0 ]; then
@@ -76,8 +79,8 @@ if [ ${stage} -le 2 ] && [ ${stop_stage} -ge 2 ]; then
     --unit_type "spm" \
     --spm_model_prefix ${bpeprefix} \
     --vocab_path="data/vocab.txt" \
-    --manifest_path="data/manifest.tiny.raw" \
-    --output_path="data/manifest.tiny"
+    --manifest_path="data/manifest.train.raw" \
+    --output_path="data/manifest.train"
 
 
     if [ $? -ne 0 ]; then

examples/librispeech/s1/local/export.sh

@@ -5,11 +5,21 @@ if [ $# != 3 ];then
     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 \
+--device ${device} \
+--nproc ${ngpu} \
 --config ${config_path} \
 --checkpoint_path ${ckpt_path_prefix} \
 --export_path ${jit_model_export_path}

examples/librispeech/s1/local/train.sh

@@ -10,6 +10,7 @@ echo "using $ngpu gpus..."
 
 config_path=$1
 ckpt_name=$2
+
 device=gpu
 if [ ngpu != 0 ];then
     device=cpu

examples/librispeech/s1/run.sh

@@ -1,20 +1,36 @@
 #!/bin/bash
 set -e
-
 source path.sh
-source ${MAIN_ROOT}/utils/parse_options.sh
 
+stage=0
+stop_stage=100
+ckpt=conformer
+avg_num=30
+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
+    bash ./local/data.sh || exit -1
+fi
 
+if [ ${stage} -le 1 ] && [ ${stop_stage} -ge 1 ]; then
     # train model, all `ckpt` under `exp` dir
-CUDA_VISIBLE_DEVICES=0 ./local/train.sh conf/conformer.yaml test
+    CUDA_VISIBLE_DEVICES=4,5,6,7 ./local/train.sh conf/conformer.yaml  ${ckpt}
+fi
 
-# test ckpt 1
-CUDA_VISIBLE_DEVICES=0 ./local/test.sh conf/conformer.yaml exp/test/checkpoints/1
+if [ ${stage} -le 2 ] && [ ${stop_stage} -ge 2 ]; then
+    # avg n best model
+    ./local/avg.sh exp/${ckpt}/checkpoints ${avg_num}
+fi
 
-# avg 1 best model
-./local/avg.sh exp/test/checkpoints 1
+if [ ${stage} -le 3 ] && [ ${stop_stage} -ge 3 ]; then
+    # test ckpt avg_n
+    CUDA_VISIBLE_DEVICES=7 ./local/test.sh conf/conformer.yaml exp/${ckpt}/checkpoints/${avg_ckpt} || exit -1
+fi
 
-# export ckpt 1
-./local/export.sh conf/conformer.yaml exp/test/checkpoints/1 exp/test/checkpoints/1.jit.model
+if [ ${stage} -le 4 ] && [ ${stop_stage} -ge 4 ]; then
+    # export ckpt avg_n
+    CUDA_VISIBLE_DEVICES= ./local/export.sh conf/conformer.yaml exp/${ckpt}/checkpoints/${avg_ckpt} exp/${ckpt}/checkpoints/${avg_ckpt}.jit
+fi

examples/tiny/s1/run.sh

@@ -1,20 +1,36 @@
 #!/bin/bash
 set -e
-
 source path.sh
+
+stage=0
+stop_stage=100
+ckpt=conformer
+avg_num=1
+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
 
+if [ ${stage} -le 1 ] && [ ${stop_stage} -ge 1 ]; then
     # train model, all `ckpt` under `exp` dir
-CUDA_VISIBLE_DEVICES=0 ./local/train.sh conf/conformer.yaml test || exit -1
+    CUDA_VISIBLE_DEVICES=0 ./local/train.sh conf/conformer.yaml  ${ckpt}
+fi
 
-# avg 1 best model
-./local/avg.sh exp/test/checkpoints 1
+if [ ${stage} -le 2 ] && [ ${stop_stage} -ge 2 ]; then
+    # avg n best model
+    ./local/avg.sh exp/${ckpt}/checkpoints ${avg_num}
+fi
 
-# test ckpt 1
-CUDA_VISIBLE_DEVICES=0 ./local/test.sh conf/conformer.yaml exp/test/checkpoints/avg_1 || exit -1
+if [ ${stage} -le 3 ] && [ ${stop_stage} -ge 3 ]; then
+    # test ckpt avg_n
+    CUDA_VISIBLE_DEVICES=0 ./local/test.sh conf/conformer.yaml exp/${ckpt}/checkpoints/${avg_ckpt} || exit -1
+fi
 
-# export ckpt 1
-CUDA_VISIBLE_DEVICES= ./local/export.sh conf/conformer.yaml exp/test/checkpoints/avg_1 exp/test/checkpoints/avg_1.jit.model
+if [ ${stage} -le 4 ] && [ ${stop_stage} -ge 4 ]; then
+    # export ckpt avg_n
+    CUDA_VISIBLE_DEVICES= ./local/export.sh conf/conformer.yaml exp/${ckpt}/checkpoints/${avg_ckpt} exp/${ckpt}/checkpoints/${avg_ckpt}.jit
+fi