pre-commit hooks

egs/mls_english/ASR/local/utils/asr_datamodule.py

@@ -36,6 +36,7 @@ from torch.utils.data import DataLoader
 
 from icefall.utils import str2bool
 
+
 class MLSEnglishHFAsrDataModule:
     """
     DataModule for MLS English ASR experiments using HuggingFace dataset.
@@ -46,6 +47,7 @@ class MLSEnglishHFAsrDataModule:
     def __init__(self, args: argparse.Namespace):
         self.args = args
         self.dataset = None
+
     #     self._validate_args()
 
     # def _validate_args(self) -> None:
@@ -123,7 +125,6 @@ class MLSEnglishHFAsrDataModule:
             help="Whether to drop last incomplete batch",
         )
 
-
         return parser
 
     def load_dataset(self, dataset_path: Optional[str] = None) -> None:
@@ -133,16 +134,17 @@ class MLSEnglishHFAsrDataModule:
 
         try:
             from datasets import load_dataset
+
             self.dataset = load_dataset(dataset_path)
             logging.info("Dataset loaded successfully")
         except ImportError:
-            raise ImportError(
+            raise ImportError("Please install datasets package: pip install datasets")
-                "Please install datasets package: pip install datasets"
-            )
         except Exception as e:
             raise RuntimeError(f"Failed to load dataset: {e}")
 
-    def _create_dataset(self, cuts: CutSet, is_train: bool = False) -> K2SpeechRecognitionDataset:
+    def _create_dataset(
+        self, cuts: CutSet, is_train: bool = False
+    ) -> K2SpeechRecognitionDataset:
         """Create appropriate dataset with transforms."""
         transforms = []
         input_transforms = []
@@ -160,9 +162,9 @@ class MLSEnglishHFAsrDataModule:
     def _create_spec_augment(self) -> SpecAugment:
         """Create SpecAugment transform based on config."""
         num_frame_masks = 10
-        num_frame_masks_parameter = inspect.signature(
+        num_frame_masks_parameter = inspect.signature(SpecAugment.__init__).parameters[
-            SpecAugment.__init__
+            "num_frame_masks"
-        ).parameters["num_frame_masks"]
+        ]
         if num_frame_masks_parameter.default == 1:
             num_frame_masks = 2
 
@@ -174,7 +176,9 @@ class MLSEnglishHFAsrDataModule:
             frames_mask_size=100,
         )
 
-    def _create_sampler(self, cuts: CutSet, shuffle: bool) -> Union[DynamicBucketingSampler, SimpleCutSampler]:
+    def _create_sampler(
+        self, cuts: CutSet, shuffle: bool
+    ) -> Union[DynamicBucketingSampler, SimpleCutSampler]:
         """Create appropriate sampler based on config."""
         if self.args.bucketing_sampler:
             return DynamicBucketingSampler(
@@ -190,7 +194,9 @@ class MLSEnglishHFAsrDataModule:
             shuffle=shuffle,
         )
 
-    def train_dataloader(self, sampler_state_dict: Optional[Dict[str, Any]] = None) -> DataLoader:
+    def train_dataloader(
+        self, sampler_state_dict: Optional[Dict[str, Any]] = None
+    ) -> DataLoader:
         """Create train dataloader."""
         cuts = self.train_cuts()
         dataset = self._create_dataset(cuts, is_train=True)
@@ -231,20 +237,17 @@ class MLSEnglishHFAsrDataModule:
     @lru_cache()
     def train_cuts(self) -> CutSet:
         return CutSet.from_huggingface_dataset(
-            self.dataset["train"], 
+            self.dataset["train"], text_key="transcript"
-            text_key="transcript"
         )
 
     @lru_cache()
     def valid_cuts(self) -> CutSet:
         return CutSet.from_huggingface_dataset(
-            self.dataset["dev"], 
+            self.dataset["dev"], text_key="transcript"
-            text_key="transcript"
         )
 
     @lru_cache()
     def test_cuts(self) -> CutSet:
         return CutSet.from_huggingface_dataset(
-            self.dataset["test"], 
+            self.dataset["test"], text_key="transcript"
-            text_key="transcript"
         )

egs/mls_english/ASR/local/utils/compute_fbank_mls_english.py

@@ -49,7 +49,7 @@ concat_params = {"gap": 1.0, "maxlen": 10.0}
 
 
 def make_cutset_blueprints(
-    mls_eng_hf_dataset_path: str = "parler-tts/mls_eng"
+    mls_eng_hf_dataset_path: str = "parler-tts/mls_eng",
 ) -> List[Tuple[str, CutSet]]:
     cut_sets = []
 
@@ -67,17 +67,14 @@ def make_cutset_blueprints(
     cut_sets.append(
         (
             "test",
-            CutSet.from_huggingface_dataset(dataset["test"], text_key="transcript")
+            CutSet.from_huggingface_dataset(dataset["test"], text_key="transcript"),
         )
     )
 
     # Create dev dataset
     logging.info("Creating dev cuts.")
     cut_sets.append(
-        (
+        ("dev", CutSet.from_huggingface_dataset(dataset["dev"], text_key="transcript"))
-            "dev",
-            CutSet.from_huggingface_dataset(dataset["dev"], text_key="transcript")
-        )
     )
 
     # Create train dataset
@@ -85,7 +82,7 @@ def make_cutset_blueprints(
     cut_sets.append(
         (
             "train",
-            CutSet.from_huggingface_dataset(dataset["train"], text_key="transcript")
+            CutSet.from_huggingface_dataset(dataset["train"], text_key="transcript"),
         )
     )
     return cut_sets

egs/mls_english/ASR/local/utils/generate_transcript.py

@@ -24,6 +24,7 @@ from typing import Optional
 from lhotse import CutSet
 from tqdm import tqdm
 
+
 def get_args():
     parser = argparse.ArgumentParser(
         description="Generate transcripts for BPE training from MLS English dataset",
@@ -53,6 +54,7 @@ def get_args():
 
     return parser.parse_args()
 
+
 def generate_transcript_from_cuts(cuts: CutSet, output_file: Path) -> None:
     """Generate transcript text file from Lhotse CutSet."""
     with open(output_file, "w") as f:
@@ -60,6 +62,7 @@ def generate_transcript_from_cuts(cuts: CutSet, output_file: Path) -> None:
             for sup in cut.supervisions:
                 f.write(f"{sup.text}\n")
 
+
 def main():
     args = get_args()
     logging.basicConfig(
@@ -73,9 +76,7 @@ def main():
     logging.info(f"Loading {args.split} split from dataset: {args.dataset_path}")
     try:
         cuts = CutSet.from_huggingface_dataset(
-            args.dataset_path,
+            args.dataset_path, split=args.split, text_key="transcript"
-            split=args.split,
-            text_key="transcript"
         )
     except Exception as e:
         logging.error(f"Failed to load dataset: {e}")
@@ -85,5 +86,6 @@ def main():
     generate_transcript_from_cuts(cuts, output_file)
     logging.info("Transcript generation completed")
 
+
 if __name__ == "__main__":
     main()

egs/mls_english/ASR/zipformer/decode.py

@@ -103,9 +103,6 @@ from pathlib import Path
 from typing import Dict, List, Optional, Tuple
 
 import k2
-# import sentencepiece as spm
-from tokenizer import Tokenizer
-
 import torch
 import torch.nn as nn
 from asr_datamodule import MLSEnglishHFAsrDataModule
@@ -123,6 +120,10 @@ from beam_search import (
     modified_beam_search_lm_shallow_fusion,
     modified_beam_search_LODR,
 )
+
+# import sentencepiece as spm
+from tokenizer import Tokenizer
+
 # from gigaspeech_scoring import asr_text_post_processing
 from train import add_model_arguments, get_model, get_params
 
@@ -384,6 +385,7 @@ def get_parser():
 
     return parser
 
+
 def asr_text_post_processing(inp):
     return inp
 
@@ -869,7 +871,6 @@ def main():
 
     sp = Tokenizer.load(Path(args.lang_dir), "bpe")  # force bpe model
 
-
     # <blk> and <unk> are defined in local/train_bpe_model.py
     params.blank_id = sp.piece_to_id("<blk>")
     params.unk_id = sp.piece_to_id("<unk>")

egs/mls_english/ASR/zipformer/train.py

@@ -1239,7 +1239,6 @@ def run(rank, world_size, args):
     # valid_dl = mls_english_corpus.valid_dataloader(valid_cuts)
     valid_dl = mls_english_corpus.valid_dataloader()
 
-
     if not params.print_diagnostics:
         scan_pessimistic_batches_for_oom(
             model=model,