update

examples/aishell/branchformer/conf/branchformer_12e_6d_2048_256.yaml

@@ -94,13 +94,16 @@ scheduler_conf:
 dataset: AudioDataset
 dataset_conf:
     index_ds: IndexDSJsonl
-    batch_sampler: DynamicBatchLocalShuffleSampler
+    batch_sampler: RankFullLocalShuffleBatchSampler
     batch_type: example # example or length
-    batch_size: 1 # if batch_type is example, batch_size is the numbers of samples; if length, batch_size is source_token_len+target_token_len;
+    batch_size: 32 # if batch_type is example, batch_size is the numbers of samples; if length, batch_size is source_token_len+target_token_len;
     max_token_length: 2048 # filter samples if source_token_len+target_token_len > max_token_length,
-    buffer_size: 500
+    buffer_size: 1024
     shuffle: True
     num_workers: 4
+    preprocessor_speech: SpeechPreprocessSpeedPerturb
+    preprocessor_speech_conf:
+      speed_perturb: [0.9, 1.0, 1.1]
 
 tokenizer: CharTokenizer
 tokenizer_conf:

examples/aishell/conformer/conf/conformer_12e_6d_2048_256.yaml

@@ -94,13 +94,16 @@ scheduler_conf:
 dataset: AudioDataset
 dataset_conf:
     index_ds: IndexDSJsonl
-    batch_sampler: DynamicBatchLocalShuffleSampler
+    batch_sampler: RankFullLocalShuffleBatchSampler
     batch_type: example # example or length
-    batch_size: 1 # if batch_type is example, batch_size is the numbers of samples; if length, batch_size is source_token_len+target_token_len;
+    batch_size: 32 # if batch_type is example, batch_size is the numbers of samples; if length, batch_size is source_token_len+target_token_len;
     max_token_length: 2048 # filter samples if source_token_len+target_token_len > max_token_length,
-    buffer_size: 500
+    buffer_size: 1024
     shuffle: True
     num_workers: 4
+    preprocessor_speech: SpeechPreprocessSpeedPerturb
+    preprocessor_speech_conf:
+      speed_perturb: [0.9, 1.0, 1.1]
 
 tokenizer: CharTokenizer
 tokenizer_conf:

examples/aishell/conformer/run.sh

@@ -109,9 +109,14 @@ if [ ${stage} -le 4 ] && [ ${stop_stage} -ge 4 ]; then
   echo "log_file: ${log_file}"
 
   gpu_num=$(echo CUDA_VISIBLE_DEVICES | awk -F "," '{print NF}')
-  torchrun \
-  --nnodes 1 \
-  --nproc_per_node ${gpu_num} \
+#  torchrun \
+#  --nnodes 1 \
+#  --nproc_per_node ${gpu_num}
+  cmd="python"
+  if [ ${gpu_num} -gt 1  ];then
+    cmd="torchrun --nnodes 1 --nproc_per_node ${gpu_num}"
+  fi
+  ${cmd} \
   ../../../funasr/bin/train.py \
   --config-path "${workspace}/conf" \
   --config-name "${config}" \

examples/aishell/e_branchformer/conf/e_branchformer_12e_6d_2048_256.yaml

@@ -94,13 +94,16 @@ scheduler_conf:
 dataset: AudioDataset
 dataset_conf:
     index_ds: IndexDSJsonl
-    batch_sampler: DynamicBatchLocalShuffleSampler
+    batch_sampler: RankFullLocalShuffleBatchSampler
     batch_type: example # example or length
-    batch_size: 1 # if batch_type is example, batch_size is the numbers of samples; if length, batch_size is source_token_len+target_token_len;
+    batch_size: 32 # if batch_type is example, batch_size is the numbers of samples; if length, batch_size is source_token_len+target_token_len;
     max_token_length: 2048 # filter samples if source_token_len+target_token_len > max_token_length,
-    buffer_size: 500
+    buffer_size: 1024
     shuffle: True
     num_workers: 4
+    preprocessor_speech: SpeechPreprocessSpeedPerturb
+    preprocessor_speech_conf:
+      speed_perturb: [0.9, 1.0, 1.1]
 
 tokenizer: CharTokenizer
 tokenizer_conf:

examples/aishell/transformer/transformer_12e_6d_2048_256.yaml → examples/aishell/transformer/conf/transformer_12e_6d_2048_256.yaml

@@ -88,13 +88,16 @@ scheduler_conf:
 dataset: AudioDataset
 dataset_conf:
     index_ds: IndexDSJsonl
-    batch_sampler: DynamicBatchLocalShuffleSampler
+    batch_sampler: RankFullLocalShuffleBatchSampler
     batch_type: example # example or length
-    batch_size: 1 # if batch_type is example, batch_size is the numbers of samples; if length, batch_size is source_token_len+target_token_len;
+    batch_size: 32 # if batch_type is example, batch_size is the numbers of samples; if length, batch_size is source_token_len+target_token_len;
     max_token_length: 2048 # filter samples if source_token_len+target_token_len > max_token_length,
-    buffer_size: 500
+    buffer_size: 1024
     shuffle: True
-    num_workers: 0
+    num_workers: 4
+    preprocessor_speech: SpeechPreprocessSpeedPerturb
+    preprocessor_speech_conf:
+      speed_perturb: [0.9, 1.0, 1.1]
 
 tokenizer: CharTokenizer
 tokenizer_conf: