update repo

egs/aishell2/conformer/run.sh

@@ -21,16 +21,16 @@ type=sound
 scp=wav.scp
 speed_perturb="0.9 1.0 1.1"
 dataset_type=large
-stage=3
-stop_stage=4
+stage=0
+stop_stage=5
 
 # feature configuration
 feats_dim=80
 nj=64
 
 # data
-tr_dir=/nfs/wangjiaming.wjm/asr_data/aishell2/AISHELL-2/iOS/data
-dev_tst_dir=/nfs/wangjiaming.wjm/asr_data/aishell2/AISHELL-DEV-TEST-SET
+tr_dir=
+dev_tst_dir=
 
 # exp tag
 tag="exp1"
@@ -107,10 +107,16 @@ if [ ${stage} -le 2 ] && [ ${stop_stage} -ge 2 ]; then
     mkdir -p ${feats_dir}/asr_stats_fbank_zh_char/${valid_set}
  fi
 
-# Training Stage
+# LM Training Stage
 world_size=$gpu_num  # run on one machine
 if [ ${stage} -le 3 ] && [ ${stop_stage} -ge 3 ]; then
-    echo "stage 3: Training"
+    echo "stage 3: LM Training"
+fi
+
+# ASR Training Stage
+world_size=$gpu_num  # run on one machine
+if [ ${stage} -le 4 ] && [ ${stop_stage} -ge 4 ]; then
+    echo "stage 4: ASR Training"
     mkdir -p ${exp_dir}/exp/${model_dir}
     mkdir -p ${exp_dir}/exp/${model_dir}/log
     INIT_FILE=${exp_dir}/exp/${model_dir}/ddp_init
@@ -151,8 +157,8 @@ if [ ${stage} -le 3 ] && [ ${stop_stage} -ge 3 ]; then
 fi
 
 # Testing Stage
-if [ ${stage} -le 4 ] && [ ${stop_stage} -ge 4 ]; then
-    echo "stage 4: Inference"
+if [ ${stage} -le 5 ] && [ ${stop_stage} -ge 5 ]; then
+    echo "stage 5: Inference"
     for dset in ${test_sets}; do
         asr_exp=${exp_dir}/exp/${model_dir}
         inference_tag="$(basename "${inference_config}" .yaml)"

egs/aishell2/data2vec_pretrain/run.sh

@@ -24,8 +24,8 @@ feats_dim=80
 nj=64
 
 # data
-tr_dir=/nfs/wangjiaming.wjm/asr_data/aishell2/AISHELL-2/iOS/data
-dev_tst_dir=/nfs/wangjiaming.wjm/asr_data/aishell2/AISHELL-DEV-TEST-SET
+tr_dir=
+dev_tst_dir=
 
 # exp tag
 tag="exp1"

egs/aishell2/paraformer/run.sh

@@ -21,16 +21,16 @@ type=sound
 scp=wav.scp
 speed_perturb="0.9 1.0 1.1"
 dataset_type=large
-stage=3
-stop_stage=4
+stage=0
+stop_stage=5
 
 # feature configuration
 feats_dim=80
 nj=64
 
 # data
-tr_dir=/nfs/wangjiaming.wjm/asr_data/aishell2/AISHELL-2/iOS/data
-dev_tst_dir=/nfs/wangjiaming.wjm/asr_data/aishell2/AISHELL-DEV-TEST-SET
+tr_dir=
+dev_tst_dir=
 
 # exp tag
 tag="exp1"
@@ -105,10 +105,16 @@ if [ ${stage} -le 2 ] && [ ${stop_stage} -ge 2 ]; then
     echo "<unk>" >> ${token_list}
  fi
 
-# Training Stage
+# LM Training Stage
 world_size=$gpu_num  # run on one machine
 if [ ${stage} -le 3 ] && [ ${stop_stage} -ge 3 ]; then
-    echo "stage 3: Training"
+    echo "stage 3: LM Training"
+fi
+
+# ASR Training Stage
+world_size=$gpu_num  # run on one machine
+if [ ${stage} -le 4 ] && [ ${stop_stage} -ge 4 ]; then
+    echo "stage 4: ASR Training"
     mkdir -p ${exp_dir}/exp/${model_dir}
     mkdir -p ${exp_dir}/exp/${model_dir}/log
     INIT_FILE=${exp_dir}/exp/${model_dir}/ddp_init
@@ -149,8 +155,8 @@ if [ ${stage} -le 3 ] && [ ${stop_stage} -ge 3 ]; then
 fi
 
 # Testing Stage
-if [ ${stage} -le 4 ] && [ ${stop_stage} -ge 4 ]; then
-    echo "stage 4: Inference"
+if [ ${stage} -le 5 ] && [ ${stop_stage} -ge 5 ]; then
+    echo "stage 5: Inference"
     for dset in ${test_sets}; do
         asr_exp=${exp_dir}/exp/${model_dir}
         inference_tag="$(basename "${inference_config}" .yaml)"

egs/aishell2/paraformerbert/run.sh

@@ -21,8 +21,8 @@ type=sound
 scp=wav.scp
 speed_perturb="0.9 1.0 1.1"
 dataset_type=large
-stage=3
-stop_stage=4
+stage=0
+stop_stage=5
 
 skip_extract_embed=false
 bert_model_name="bert-base-chinese"
@@ -32,8 +32,8 @@ feats_dim=80
 nj=64
 
 # data
-tr_dir=/nfs/wangjiaming.wjm/asr_data/aishell2/AISHELL-2/iOS/data
-dev_tst_dir=/nfs/wangjiaming.wjm/asr_data/aishell2/AISHELL-DEV-TEST-SET
+tr_dir=
+dev_tst_dir=
 
 # exp tag
 tag="exp1"
@@ -108,10 +108,16 @@ if [ ${stage} -le 2 ] && [ ${stop_stage} -ge 2 ]; then
     echo "<unk>" >> ${token_list}
  fi
 
-# Training Stage
+# LM Training Stage
 world_size=$gpu_num  # run on one machine
 if [ ${stage} -le 3 ] && [ ${stop_stage} -ge 3 ]; then
-    echo "stage 3: Training"
+    echo "stage 3: LM Training"
+fi
+
+# ASR Training Stage
+world_size=$gpu_num  # run on one machine
+if [ ${stage} -le 4 ] && [ ${stop_stage} -ge 4 ]; then
+    echo "stage 4: ASR Training"
     if ! "${skip_extract_embed}"; then
         echo "extract embeddings..."
         local/extract_embeds.sh \
@@ -160,8 +166,8 @@ if [ ${stage} -le 3 ] && [ ${stop_stage} -ge 3 ]; then
 fi
 
 # Testing Stage
-if [ ${stage} -le 4 ] && [ ${stop_stage} -ge 4 ]; then
-    echo "stage 4: Inference"
+if [ ${stage} -le 5 ] && [ ${stop_stage} -ge 5 ]; then
+    echo "stage 5: Inference"
     for dset in ${test_sets}; do
         asr_exp=${exp_dir}/exp/${model_dir}
         inference_tag="$(basename "${inference_config}" .yaml)"

egs/aishell2/transformer/run.sh

@@ -21,16 +21,16 @@ type=sound
 scp=wav.scp
 speed_perturb="0.9 1.0 1.1"
 dataset_type=large
-stage=3
-stop_stage=4
+stage=0
+stop_stage=5
 
 # feature configuration
 feats_dim=80
 nj=64
 
 # data
-tr_dir=/nfs/wangjiaming.wjm/asr_data/aishell2/AISHELL-2/iOS/data
-dev_tst_dir=/nfs/wangjiaming.wjm/asr_data/aishell2/AISHELL-DEV-TEST-SET
+tr_dir=
+dev_tst_dir=
 
 # exp tag
 tag="exp1"
@@ -105,10 +105,16 @@ if [ ${stage} -le 2 ] && [ ${stop_stage} -ge 2 ]; then
     echo "<unk>" >> ${token_list}
  fi
 
-# Training Stage
+# LM Training Stage
 world_size=$gpu_num  # run on one machine
 if [ ${stage} -le 3 ] && [ ${stop_stage} -ge 3 ]; then
-    echo "stage 3: Training"
+    echo "stage 3: LM Training"
+fi
+
+# ASR Training Stage
+world_size=$gpu_num  # run on one machine
+if [ ${stage} -le 4 ] && [ ${stop_stage} -ge 4 ]; then
+    echo "stage 4: ASR Training"
     mkdir -p ${exp_dir}/exp/${model_dir}
     mkdir -p ${exp_dir}/exp/${model_dir}/log
     INIT_FILE=${exp_dir}/exp/${model_dir}/ddp_init
@@ -149,8 +155,8 @@ if [ ${stage} -le 3 ] && [ ${stop_stage} -ge 3 ]; then
 fi
 
 # Testing Stage
-if [ ${stage} -le 4 ] && [ ${stop_stage} -ge 4 ]; then
-    echo "stage 4: Inference"
+if [ ${stage} -le 5 ] && [ ${stop_stage} -ge 5 ]; then
+    echo "stage 5 Inference"
     for dset in ${test_sets}; do
         asr_exp=${exp_dir}/exp/${model_dir}
         inference_tag="$(basename "${inference_config}" .yaml)"

funasr/models/e2e_asr_transducer.py

@@ -17,7 +17,7 @@ from funasr.models.joint_net.joint_network import JointNetwork
 from funasr.modules.nets_utils import get_transducer_task_io
 from funasr.layers.abs_normalize import AbsNormalize
 from funasr.torch_utils.device_funcs import force_gatherable
-from funasr.train.abs_espnet_model import AbsESPnetModel
+from funasr.models.base_model import FunASRModel
 
 if V(torch.__version__) >= V("1.6.0"):
     from torch.cuda.amp import autocast
@@ -28,7 +28,7 @@ else:
         yield
 
 
-class TransducerModel(AbsESPnetModel):
+class TransducerModel(FunASRModel):
     """ESPnet2ASRTransducerModel module definition.
 
     Args:

funasr/models/e2e_sa_asr.py

@@ -29,7 +29,7 @@ from funasr.modules.add_sos_eos import add_sos_eos
 from funasr.modules.e2e_asr_common import ErrorCalculator
 from funasr.modules.nets_utils import th_accuracy
 from funasr.torch_utils.device_funcs import force_gatherable
-from funasr.train.abs_espnet_model import AbsESPnetModel
+from funasr.models.base_model import FunASRModel
 
 if LooseVersion(torch.__version__) >= LooseVersion("1.6.0"):
     from torch.cuda.amp import autocast
@@ -40,7 +40,7 @@ else:
         yield
 
 
-class ESPnetASRModel(AbsESPnetModel):
+class ESPnetASRModel(FunASRModel):
     """CTC-attention hybrid Encoder-Decoder model"""
 
     def __init__(

funasr/tasks/sa_asr.py

@@ -70,11 +70,11 @@ from funasr.models.preencoder.sinc import LightweightSincConvs
 from funasr.models.specaug.abs_specaug import AbsSpecAug
 from funasr.models.specaug.specaug import SpecAug
 from funasr.models.specaug.specaug import SpecAugLFR
+from funasr.models.base_model import FunASRModel
 from funasr.modules.subsampling import Conv1dSubsampling
 from funasr.tasks.abs_task import AbsTask
 from funasr.text.phoneme_tokenizer import g2p_choices
 from funasr.torch_utils.initialize import initialize
-from funasr.train.abs_espnet_model import AbsESPnetModel
 from funasr.train.class_choices import ClassChoices
 from funasr.train.trainer import Trainer
 from funasr.utils.get_default_kwargs import get_default_kwargs
@@ -129,7 +129,7 @@ model_choices = ClassChoices(
         mfcca=MFCCA,
         timestamp_prediction=TimestampPredictor,
     ),
-    type_check=AbsESPnetModel,
+    type_check=FunASRModel,
     default="asr",
 )
 preencoder_choices = ClassChoices(