add the whole of aishell asr1

examples/aishell/asr1/conf/chunk_conformer.yaml

@@ -1,5 +1,6 @@
-# network architecture
-model:
+############################################
+#           Network Architecture           #
+############################################
 cmvn_file: 
 cmvn_file_type: "json"
 # encoder related
@@ -23,7 +24,6 @@ model:
     use_dynamic_chunk: true
     cnn_module_norm: 'layer_norm' # using nn.LayerNorm makes model converge faster
     use_dynamic_left_chunk: false
-
 # decoder related
 decoder: transformer
 decoder_conf:
@@ -34,21 +34,25 @@ model:
     positional_dropout_rate: 0.1
     self_attention_dropout_rate: 0.0
     src_attention_dropout_rate: 0.0
-
 # hybrid CTC/attention
 model_conf:
     ctc_weight: 0.3
     lsm_weight: 0.1     # label smoothing option
     length_normalized_loss: false
 
+###########################################
+#                   Data                  #
+###########################################
 
-data:
 train_manifest: data/manifest.train
 dev_manifest: data/manifest.dev
 test_manifest: data/manifest.test
 
 
-collator:
+###########################################
+#              Dataloader                 #
+###########################################
+
 vocab_filepath: data/lang_char/vocab.txt 
 unit_type: 'char'
 augmentation_config: conf/preprocess.yaml
@@ -69,8 +73,9 @@ collator:
 subsampling_factor: 1
 num_encs: 1
 
-
-training:
+###########################################
+#                 training                #
+###########################################
 n_epoch: 240 
 accum_grad: 2
 global_grad_clip: 5.0
@@ -87,17 +92,3 @@ training:
   kbest_n: 50
   latest_n: 5
 
-
-decoding:
-  beam_size: 10
-  batch_size: 128
-  error_rate_type: cer 
-  decoding_method: attention # 'attention', 'ctc_greedy_search', 'ctc_prefix_beam_search', 'attention_rescoring'
-  ctc_weight: 0.5 # ctc weight for attention rescoring decode mode.
-  decoding_chunk_size: -1 # decoding chunk size. Defaults to -1.
-      # <0: for decoding, use full chunk.
-      # >0: for decoding, use fixed chunk size as set.
-      # 0: used for training, it's prohibited here. 
-  num_decoding_left_chunks: -1  # number of left chunks for decoding. Defaults to -1.
-  simulate_streaming: False  # simulate streaming inference. Defaults to False.
-

examples/aishell/asr1/conf/conformer.yaml

@@ -1,7 +1,6 @@
 ############################################
 #           Network Architecture           #
 ############################################
-#model:
 cmvn_file: 
 cmvn_file_type: "json"
 # encoder related
@@ -42,7 +41,6 @@ model_conf:
 ###########################################
 #                   Data                  #
 ###########################################
-#data:
 train_manifest: data/manifest.train
 dev_manifest: data/manifest.dev
 test_manifest: data/manifest.test
@@ -50,11 +48,9 @@ test_manifest: data/manifest.test
 ###########################################
 #              Dataloader                 #
 ###########################################
-#collator:
 vocab_filepath: data/lang_char/vocab.txt 
 unit_type: 'char'
 augmentation_config: conf/preprocess.yaml
-spm_model_prefix: ''
 feat_dim: 80
 stride_ms: 10.0
 window_ms: 25.0
@@ -75,7 +71,6 @@ num_encs: 1
 ###########################################
 #                 training                #
 ###########################################
-#training:
 n_epoch: 240 
 accum_grad: 2
 global_grad_clip: 5.0

examples/aishell/asr1/conf/transformer.yaml

@@ -1,5 +1,6 @@
-# network architecture
-model:
+############################################
+#           Network Architecture           #
+############################################
 cmvn_file: 
 cmvn_file_type: "json"
 # encoder related
@@ -14,7 +15,6 @@ model:
     attention_dropout_rate: 0.0
     input_layer: conv2d # encoder input type, you can chose conv2d, conv2d6 and conv2d8
     normalize_before: true
-
 # decoder related
 decoder: transformer
 decoder_conf:
@@ -32,15 +32,18 @@ model:
     lsm_weight: 0.1     # label smoothing option
     length_normalized_loss: false
 
-
+###########################################
+#                   Data                  #
+###########################################
 # https://yaml.org/type/float.html
-data:
 train_manifest: data/manifest.train
 dev_manifest: data/manifest.dev
 test_manifest: data/manifest.test
 
 
-collator:
+###########################################
+#              Dataloader                 #
+###########################################
 unit_type: 'char'
 vocab_filepath: data/lang_char/vocab.txt 
 feat_dim: 80
@@ -61,9 +64,9 @@ collator:
 subsampling_factor: 1
 num_encs: 1
 
-
-
-training:
+###########################################
+#                 training                #
+###########################################
 n_epoch: 240 
 accum_grad: 2
 global_grad_clip: 5.0
@@ -79,17 +82,3 @@ training:
 checkpoint:
   kbest_n: 50
   latest_n: 5
-
-
-decoding:
-  beam_size: 10
-  batch_size: 128
-  error_rate_type: cer 
-  decoding_method: attention # 'attention', 'ctc_greedy_search', 'ctc_prefix_beam_search', 'attention_rescoring'
-  ctc_weight: 0.5 # ctc weight for attention rescoring decode mode.
-  decoding_chunk_size: -1 # decoding chunk size. Defaults to -1.
-      # <0: for decoding, use full chunk.
-      # >0: for decoding, use fixed chunk size as set.
-      # 0: used for training, it's prohibited here. 
-  num_decoding_left_chunks: -1  # number of left chunks for decoding. Defaults to -1.
-  simulate_streaming: False  # simulate streaming inference. Defaults to False.

examples/aishell/asr1/conf/tuning/chunk_decode.yaml

@@ -0,0 +1,11 @@
+beam_size: 10
+decode_batch_size: 128
+error_rate_type: cer 
+decoding_method: attention # 'attention', 'ctc_greedy_search', 'ctc_prefix_beam_search', 'attention_rescoring'
+ctc_weight: 0.5 # ctc weight for attention rescoring decode mode.
+decoding_chunk_size: -1 # decoding chunk size. Defaults to -1.
+    # <0: for decoding, use full chunk.
+    # >0: for decoding, use fixed chunk size as set.
+    # 0: used for training, it's prohibited here. 
+num_decoding_left_chunks: -1  # number of left chunks for decoding. Defaults to -1.
+simulate_streaming: False  # simulate streaming inference. Defaults to False.

examples/aishell/asr1/conf/tuning/decode.yaml

@@ -1,4 +1,3 @@
-#decoding:
 beam_size: 10
 decode_batch_size: 128
 error_rate_type: cer 

examples/aishell/asr1/run.sh

@@ -6,7 +6,7 @@ gpus=0,1,2,3
 stage=0
 stop_stage=50
 conf_path=conf/conformer.yaml
-decode_conf_path=conf/decode.yaml
+decode_conf_path=conf/tuning/decode.yaml
 avg_num=20
 audio_file=data/demo_01_03.wav