Fix typos, remove unused packages, normalize comments (#1678)

egs/librispeech/ASR/tdnn_lstm_ctc/asr_datamodule.py

@@ -1,4 +1,4 @@
-#  Copyright      2021  Piotr Żelasko
+# Copyright      2021  Piotr Żelasko
 # Copyright      2022  Xiaomi Corporation     (Author: Mingshuang Luo)
 #
 # See ../../../../LICENSE for clarification regarding multiple authors

egs/librispeech/ASR/zipformer/decode.py

@@ -133,7 +133,6 @@ from icefall.checkpoint import (
 from icefall.lexicon import Lexicon
 from icefall.utils import (
     AttributeDict,
-    make_pad_mask,
     setup_logger,
     store_transcripts,
     str2bool,

egs/librispeech/ASR/zipformer/export-onnx-ctc.py

@@ -72,7 +72,6 @@ import k2
 import onnx
 import torch
 import torch.nn as nn
-from decoder import Decoder
 from onnxruntime.quantization import QuantType, quantize_dynamic
 from scaling_converter import convert_scaled_to_non_scaled
 from train import add_model_arguments, get_model, get_params

egs/librispeech/ASR/zipformer/jit_pretrained_streaming.py

@@ -40,15 +40,12 @@ Usage of this script:
 
 import argparse
 import logging
-import math
 from typing import List, Optional
 
 import k2
-import kaldifeat
 import torch
 import torchaudio
 from kaldifeat import FbankOptions, OnlineFbank, OnlineFeature
-from torch.nn.utils.rnn import pad_sequence
 
 
 def get_parser():

egs/librispeech/ASR/zipformer/joiner.py

@@ -45,7 +45,7 @@ class Joiner(nn.Module):
             Output from the encoder. Its shape is (N, T, s_range, C).
           decoder_out:
             Output from the decoder. Its shape is (N, T, s_range, C).
-           project_input:
+          project_input:
             If true, apply input projections encoder_proj and decoder_proj.
             If this is false, it is the user's responsibility to do this
             manually.

egs/librispeech/ASR/zipformer/onnx_check.py

@@ -82,8 +82,6 @@ import logging
 import torch
 from onnx_pretrained import OnnxModel
 
-from icefall import is_module_available
-
 
 def get_parser():
     parser = argparse.ArgumentParser(

egs/librispeech/ASR/zipformer/optim.py

@@ -1,4 +1,4 @@
-#      Copyright      2022  Xiaomi Corp.        (authors: Daniel Povey)
+# Copyright      2022  Xiaomi Corp.        (authors: Daniel Povey)
 #
 # See ../LICENSE for clarification regarding multiple authors
 #
@@ -22,7 +22,7 @@ from typing import Dict, List, Optional, Tuple, Union
 
 import torch
 from lhotse.utils import fix_random_seed
-from torch import Tensor, nn
+from torch import Tensor
 from torch.optim import Optimizer
 
 

egs/librispeech/ASR/zipformer/pretrained.py

@@ -126,8 +126,6 @@ from export import num_tokens
 from torch.nn.utils.rnn import pad_sequence
 from train import add_model_arguments, get_model, get_params
 
-from icefall.utils import make_pad_mask
-
 
 def get_parser():
     parser = argparse.ArgumentParser(

egs/librispeech/ASR/zipformer/scaling_converter.py

@@ -1,4 +1,5 @@
-# Copyright    2022-2023  Xiaomi Corp.        (authors: Fangjun Kuang, Zengwei Yao)
+# Copyright    2022-2023  Xiaomi Corp.        (authors: Fangjun Kuang,
+#                                                       Zengwei Yao)
 #
 # See ../../../../LICENSE for clarification regarding multiple authors
 #
@@ -22,7 +23,7 @@ BasicNorm is replaced by a module with `exp` removed.
 """
 
 import copy
-from typing import List, Tuple
+from typing import List
 
 import torch
 import torch.nn as nn

egs/librispeech/ASR/zipformer/train.py

@@ -512,10 +512,6 @@ def get_params() -> AttributeDict:
 
         - subsampling_factor:  The subsampling factor for the model.
 
-        - encoder_dim: Hidden dim for multi-head attention model.
-
-        - num_decoder_layers: Number of decoder layer of transformer decoder.
-
         - warm_step: The warmup period that dictates the decay of the
               scale on "simple" (un-pruned) loss.
     """