Refactoring.

egs/librispeech/ASR/transducer_stateless/decode.py

@@ -56,13 +56,9 @@ import torch
 import torch.nn as nn
 from asr_datamodule import LibriSpeechAsrDataModule
 from beam_search import beam_search, greedy_search, modified_beam_search
-from conformer import Conformer
-from decoder import Decoder
-from joiner import Joiner
-from model import Transducer
+from train import get_transducer_model, get_params
 
 from icefall.checkpoint import average_checkpoints, load_checkpoint
-from icefall.env import get_env_info
 from icefall.utils import (
     AttributeDict,
     setup_logger,
@@ -143,70 +139,6 @@ def get_parser():
     return parser
 
 
-def get_params() -> AttributeDict:
-    params = AttributeDict(
-        {
-            # parameters for conformer
-            "feature_dim": 80,
-            "encoder_out_dim": 512,
-            "subsampling_factor": 4,
-            "attention_dim": 512,
-            "nhead": 8,
-            "dim_feedforward": 2048,
-            "num_encoder_layers": 12,
-            "vgg_frontend": False,
-            "env_info": get_env_info(),
-        }
-    )
-    return params
-
-
-def get_encoder_model(params: AttributeDict):
-    # TODO: We can add an option to switch between Conformer and Transformer
-    encoder = Conformer(
-        num_features=params.feature_dim,
-        output_dim=params.encoder_out_dim,
-        subsampling_factor=params.subsampling_factor,
-        d_model=params.attention_dim,
-        nhead=params.nhead,
-        dim_feedforward=params.dim_feedforward,
-        num_encoder_layers=params.num_encoder_layers,
-        vgg_frontend=params.vgg_frontend,
-    )
-    return encoder
-
-
-def get_decoder_model(params: AttributeDict):
-    decoder = Decoder(
-        vocab_size=params.vocab_size,
-        embedding_dim=params.encoder_out_dim,
-        blank_id=params.blank_id,
-        context_size=params.context_size,
-    )
-    return decoder
-
-
-def get_joiner_model(params: AttributeDict):
-    joiner = Joiner(
-        input_dim=params.encoder_out_dim,
-        output_dim=params.vocab_size,
-    )
-    return joiner
-
-
-def get_transducer_model(params: AttributeDict):
-    encoder = get_encoder_model(params)
-    decoder = get_decoder_model(params)
-    joiner = get_joiner_model(params)
-
-    model = Transducer(
-        encoder=encoder,
-        decoder=decoder,
-        joiner=joiner,
-    )
-    return model
-
-
 def decode_one_batch(
     params: AttributeDict,
     model: nn.Module,
@@ -487,8 +419,5 @@ def main():
     logging.info("Done!")
 
 
-torch.set_num_threads(1)
-torch.set_num_interop_threads(1)
-
 if __name__ == "__main__":
     main()