fix lint

egs/multi_zh-hans/ASR/whisper/decode.py

@@ -57,8 +57,8 @@
 from multi_dataset import MultiDataset
 from tn.chinese.normalizer import Normalizer
 from whisper.normalizers import BasicTextNormalizer
-from whisper_encoder_forward_monkey_patch import replace_whisper_encoder_forward
 from whisper_decoder_forward_monkey_patch import replace_whisper_decoder_forward
+from whisper_encoder_forward_monkey_patch import replace_whisper_encoder_forward
 from zhconv import convert
 
 from icefall.checkpoint import average_checkpoints_with_averaged_model, load_checkpoint
@@ -297,6 +297,7 @@ def decode_one_batch(
     print(hyps)
     return {"beam-search": hyps}
 
+
 def decode_dataset(
     dl: torch.utils.data.DataLoader,
     params: AttributeDict,
@@ -314,6 +315,7 @@ def decode_dataset(
     Returns:
         Return a dict, whose key may be "beam-search".
     """
+
     def normalize_text_alimeeting(text: str, normalize: str = "m2met") -> str:
         """
         Text normalization similar to M2MeT challenge baseline.
@@ -323,6 +325,7 @@ def normalize_text_alimeeting(text: str, normalize: str = "m2met") -> str:
             return text
         elif normalize == "m2met":
             import re
+
             text = text.replace(" ", "")
             text = text.replace("<sil>", "")
             text = text.replace("<%>", "")
@@ -348,6 +351,7 @@ def normalize_text_alimeeting(text: str, normalize: str = "m2met") -> str:
             text = text.replace("、", "")
             text = text.replace("？", "")
             return text
+
     results = []
 
     num_cuts = 0

egs/multi_zh-hans/ASR/whisper/multi_dataset.py

@@ -244,4 +244,4 @@ def test_cuts(self) -> Dict[str, CutSet]:
             # "kespeech-asr_dev_phase2": kespeech_dev_phase2_cuts,
             # "wenetspeech-net_test": wenetspeech_test_net_cuts,
             # "wenetspeech_dev": wenetspeech_dev_cuts,
-        }
+        }

egs/multi_zh-hans/ASR/whisper/train.py

@@ -65,8 +65,8 @@
 from torch.nn.functional import pad as pad_tensor
 from torch.nn.parallel import DistributedDataParallel as DDP
 from torch.utils.tensorboard import SummaryWriter
-from whisper_encoder_forward_monkey_patch import replace_whisper_encoder_forward
 from whisper_decoder_forward_monkey_patch import replace_whisper_decoder_forward
+from whisper_encoder_forward_monkey_patch import replace_whisper_encoder_forward
 
 from icefall import diagnostics
 from icefall.checkpoint import load_checkpoint, remove_checkpoints
@@ -458,6 +458,7 @@ def normalize_text_alimeeting(text: str, normalize: str = "m2met") -> str:
             return text
         elif normalize == "m2met":
             import re
+
             text = text.replace(" ", "")
             text = text.replace("<sil>", "")
             text = text.replace("<%>", "")

egs/multi_zh-hans/ASR/whisper/whisper_decoder_forward_monkey_patch.py

@@ -1,11 +1,12 @@
+from typing import Dict, Iterable, Optional
+
+import numpy as np
 import torch
 import torch.nn.functional as F
 import whisper
-from torch import Tensor
-from torch import nn
-from typing import Dict, Iterable, Optional
-from whisper.model import ResidualAttentionBlock, LayerNorm
-import numpy as np
+from torch import Tensor, nn
+from whisper.model import LayerNorm, ResidualAttentionBlock
+
 
 def forward(self, x: Tensor, xa: Tensor, kv_cache: Optional[dict] = None):
     """
@@ -19,10 +20,7 @@ def forward(self, x: Tensor, xa: Tensor, kv_cache: Optional[dict] = None):
         self.token_embedding(x)
         + self.positional_embedding[offset : offset + x.shape[-1]]
     )
-    x = (
-        x
-        + self.positional_embedding[offset : offset + x.shape[1]]
-    )
+    x = x + self.positional_embedding[offset : offset + x.shape[1]]
     x = x.to(xa.dtype)
 
     # for block in self.blocks:
@@ -39,6 +37,7 @@ def forward(self, x: Tensor, xa: Tensor, kv_cache: Optional[dict] = None):
 
     return logits
 
+
 def replace_whisper_decoder_forward():
     """
     This function monkey patches the forward method of the whisper encoder.

egs/speechio/ASR/RESULTS.md

@@ -20,7 +20,7 @@
 | 10 | **whisper-large-ft-v0**  | **6.34%** | 2023.03 |
 | 11 | baidu_pro_api_zh | 7.29% | 2023.12 |
 
-Note: Above API results are from [SPEECHIO](https://github.com/SpeechColab/Leaderboard). All results used the default [normalize method.](https://github.com/SpeechColab/Leaderboard/blob/master/utils/benchmark.sh#L67) 
+Note: Above API results are from [SPEECHIO](https://github.com/SpeechColab/Leaderboard). All results used the default [normalize method.](https://github.com/SpeechColab/Leaderboard/blob/master/utils/benchmark.sh#L67)
 
 <details><summary> Detail all models </summary><p>
 

egs/speechio/ASR/local/normalize_results.py

@@ -29,9 +29,10 @@
 from typing import Dict, List, Optional, Tuple
 
 import kaldialign
+from speechio_norm import TextNorm
 
 from icefall.utils import store_transcripts, write_error_stats
-from speechio_norm import TextNorm
+
 
 def get_parser():
     parser = argparse.ArgumentParser(
@@ -140,6 +141,7 @@ def get_filenames(
         results.append(whisper_filename)
     return results
 
+
 def main():
     parser = get_parser()
     args = parser.parse_args()