fix egs

examples/aishell/local/infer.sh

@@ -1,5 +1,9 @@
 #! /usr/bin/env bash
 
+if [[ $# != 1 ]];
+    echo "usage: $0 ckpt-path"
+    exit -1
+fi
 
 # download language model
 bash local/download_lm_ch.sh

examples/aishell/local/infer_golden.sh

@@ -1,31 +0,0 @@
-#! /usr/bin/env bash
-
-# download language model
-bash local/download_lm_ch.sh
-if [ $? -ne 0 ]; then
-    exit 1
-fi
-
-# download well-trained model
-bash local/download_model.sh
-if [ $? -ne 0 ]; then
-    exit 1
-fi
-
-# infer
-CUDA_VISIBLE_DEVICES=0 \
-python3 -u ${BIN_DIR}/infer.py \
---device 'gpu' \
---nproc 1 \
---config conf/deepspeech2.yaml \
---checkpoint_path data/pretrain/params.pdparams  \
---opts data.mean_std_filepath data/pretrain/mean_std.npz  \
---opts data.vocab_filepath data/pretrain/vocab.txt
-
-if [ $? -ne 0 ]; then
-    echo "Failed in inference!"
-    exit 1
-fi
-
-
-exit 0

examples/aishell/local/server.sh

@@ -2,7 +2,7 @@
 # TODO: replace the model with a mandarin model
 
 if [[ $# != 1 ]];then
-   echo "usage: server.sh checkpoint_path"
+   echo "usage: $1 checkpoint_path"
    exit -1
 fi
 

examples/aishell/local/test.sh

@@ -10,7 +10,7 @@ python3 -u ${BIN_DIR}/test.py \
 --device 'gpu' \
 --nproc 1 \
 --config conf/deepspeech2.yaml \
---checkpoint_path ${1} 
+--output ckpt
 
 if [ $? -ne 0 ]; then
     echo "Failed in evaluation!"

examples/aishell/local/test_golden.sh

@@ -1,31 +0,0 @@
-#! /usr/bin/env bash
-
-# download language model
-bash local/download_lm_ch.sh
-if [ $? -ne 0 ]; then
-    exit 1
-fi
-
-# download well-trained model
-bash local/download_model.sh
-if [ $? -ne 0 ]; then
-    exit 1
-fi
-
-# evaluate model
-CUDA_VISIBLE_DEVICES=0 \
-python3 -u ${BIN_DIR}/test.py \
---device 'gpu' \
---nproc 1 \
---config conf/deepspeech2.yaml \
---checkpoint_path data/pretrain/params.pdparams  \
---opts data.mean_std_filepath data/pretrain/mean_std.npz  \
---opts data.vocab_filepath data/pretrain/vocab.txt
-
-if [ $? -ne 0 ]; then
-    echo "Failed in evaluation!"
-    exit 1
-fi
-
-
-exit 0

examples/aishell/local/train.sh

@@ -11,7 +11,7 @@ python3 -u ${BIN_DIR}/train.py \
 --device 'gpu' \
 --nproc ${ngpu} \
 --config conf/deepspeech2.yaml \
---output ckpt
+--output ckpt-${1}
 
 
 if [ $? -ne 0 ]; then

examples/aishell/run.sh

@@ -10,7 +10,10 @@ bash ./local/data.sh
 CUDA_VISIBLE_DEVICES=0,1,2,3 bash ./local/train.sh
 
 # test model
-CUDA_VISIBLE_DEVICES=0 bash ./local/test.sh ckpt/checkpoints/step-3284
+CUDA_VISIBLE_DEVICES=0 bash ./local/test.sh
 
 # infer model
 CUDA_VISIBLE_DEVICES=0 bash ./local/infer.sh ckpt/checkpoints/step-3284
+
+# export model
+bash ./local/export.sh ckpt/checkpoints/step-3284 jit.model

examples/librispeech/local/infer.sh

@@ -1,5 +1,10 @@
 #! /usr/bin/env bash
 
+if [[ $# != 1 ]];
+    echo "usage: $0 ckpt-path"
+    exit -1
+fi
+
 # download language model
 bash local/download_lm_en.sh
 if [ $? -ne 0 ]; then

examples/librispeech/run.sh

@@ -7,10 +7,13 @@ source path.sh
 bash ./local/data.sh
 
 # train model
-CUDA_VISIBLE_DEVICES=0,1,2,3,4,5,6,7 bash ./local/train.sh
+CUDA_VISIBLE_DEVICES=0,1,2,3 bash ./local/train.sh
 
 # test model
 CUDA_VISIBLE_DEVICES=0  bash ./local/test.sh
 
 # infer model
-CUDA_VISIBLE_DEVICES=0 bash ./local/infer.sh
+CUDA_VISIBLE_DEVICES=0  bash ./local/infer.sh ckpt/checkpoints/step-3284
+
+# export model
+bash ./local/export.sh ckpt/checkpoints/step-3284 jit.model

examples/tiny/README.md

@@ -1,7 +1,8 @@
 # Tiny Example
 
 1. `source path.sh`
-2. `bash run.sh`
+3. set `CUDA_VISIBLE_DEVICES` as you need.
+2. demo scrpt is `bash run.sh`. You can run commond separately as needed.
 
 ## Steps
 - Prepare the data
@@ -26,11 +27,7 @@
     bash local/infer.sh
     ```
 
-    `infer.sh` will show us some speech-to-text decoding results for several (default: 10) samples with the trained model. The performance might not be good now as the current model is only trained with a toy subset of LibriSpeech. To see the results with a better model, you can download a well-trained (trained for several days, with the complete LibriSpeech) model and do the inference:
+    `infer.sh` will show us some speech-to-text decoding results for several (default: 10) samples with the trained model. The performance might not be good now as the current model is only trained with a toy subset of LibriSpeech. To see the results with a better model, you can download a well-trained (trained for several days, with the complete LibriSpeech) model and do the inference.
-
-    ```bash
-    bash local/infer_golden.sh
-    ```
 
 - Evaluate an existing model
 
@@ -40,6 +37,15 @@
 
     `test.sh` will evaluate the model with Word Error Rate (or Character Error Rate) measurement. Similarly, you can also download a well-trained model and test its performance:
 
+
+- Export jit model
+
+    ```bash
+    bash local/export.sh ckpt_path saved_jit_model_path
+    ```
+
+- Tune hyper paerameter
+
     ```bash
-    bash local/test_golden.sh
+    bash local/tune.sh
     ```

examples/tiny/local/infer.sh

@@ -1,17 +1,21 @@
 #! /usr/bin/env bash
 
+if [[ $# != 1 ]];
+    echo "usage: $0 ckpt-path"
+    exit -1
+fi
+
 # download language model
 bash local/download_lm_en.sh
 if [ $? -ne 0 ]; then
     exit 1
 fi
 
-CUDA_VISIBLE_DEVICES=0 \
 python3 -u ${BIN_DIR}/infer.py \
 --device 'gpu' \
 --nproc 1 \
 --config conf/deepspeech2.yaml \
---output ckpt
+--checkpoint_path ${1} 
 
 
 if [ $? -ne 0 ]; then