formatted the entire LibriSpeech recipe (#1270)

* formatted the entire librispeech recipe

* minor updates

egs/librispeech/ASR/conformer_ctc/train.py

@@ -557,7 +557,6 @@ def train_one_epoch(
             )
 
         if batch_idx % params.log_interval == 0:
-
             if tb_writer is not None:
                 loss_info.write_summary(
                     tb_writer, "train/current_", params.batch_idx_train

egs/librispeech/ASR/local/download_lm.py

@@ -43,6 +43,7 @@
 
 from tqdm.auto import tqdm
 
+
 # This function is copied from lhotse
 def tqdm_urlretrieve_hook(t):
     """Wraps tqdm instance.

egs/librispeech/ASR/long_file_recog/beam_search.py

@@ -236,7 +236,7 @@ def greedy_search_batch(
     encoder_out = model.joiner.encoder_proj(packed_encoder_out.data)
 
     offset = 0
-    for (t, batch_size) in enumerate(batch_size_list):
+    for t, batch_size in enumerate(batch_size_list):
         start = offset
         end = offset + batch_size
         current_encoder_out = encoder_out.data[start:end]
@@ -507,7 +507,7 @@ def modified_beam_search(
 
     offset = 0
     finalized_B = []
-    for (t, batch_size) in enumerate(batch_size_list):
+    for t, batch_size in enumerate(batch_size_list):
         start = offset
         end = offset + batch_size
         current_encoder_out = encoder_out.data[start:end]

egs/librispeech/ASR/long_file_recog/merge_chunks.py

@@ -162,7 +162,6 @@ def _merge(cut_list: List[Cut], rec_id: str, utt_idx: int):
 
     futures = []
     with ThreadPoolExecutor(max_workers=1) as executor:
-
         for cut in cuts_chunk:
             cur_rec_id = cut.recording.id
             if len(cut_list) == 0:

egs/librispeech/ASR/long_file_recog/recognize.py

@@ -264,6 +264,7 @@ def decode_dataset(
         - timestamps of reference transcript
         - timestamps of predicted result
     """
+
     #  Background worker to add alignemnt and save cuts to disk.
     def _save_worker(
         cuts: List[Cut],

egs/librispeech/ASR/pruned2_knowledge/optim.py

@@ -66,7 +66,6 @@ def __init__(
         weight_decay=1e-3,
         target_rms=0.1,
     ):
-
         if not 0.0 <= lr:
             raise ValueError("Invalid learning rate: {}".format(lr))
         if not 0.0 <= eps:

egs/librispeech/ASR/pruned_transducer_stateless2/beam_search.py

@@ -719,7 +719,7 @@ def greedy_search_batch(
     encoder_out = model.joiner.encoder_proj(packed_encoder_out.data)
 
     offset = 0
-    for (t, batch_size) in enumerate(batch_size_list):
+    for t, batch_size in enumerate(batch_size_list):
         start = offset
         end = offset + batch_size
         current_encoder_out = encoder_out.data[start:end]
@@ -1019,7 +1019,7 @@ def modified_beam_search(
 
     offset = 0
     finalized_B = []
-    for (t, batch_size) in enumerate(batch_size_list):
+    for t, batch_size in enumerate(batch_size_list):
         start = offset
         end = offset + batch_size
         current_encoder_out = encoder_out.data[start:end]
@@ -1227,7 +1227,7 @@ def modified_beam_search_lm_rescore(
 
     offset = 0
     finalized_B = []
-    for (t, batch_size) in enumerate(batch_size_list):
+    for t, batch_size in enumerate(batch_size_list):
         start = offset
         end = offset + batch_size
         current_encoder_out = encoder_out.data[start:end]
@@ -1427,7 +1427,7 @@ def modified_beam_search_lm_rescore_LODR(
 
     offset = 0
     finalized_B = []
-    for (t, batch_size) in enumerate(batch_size_list):
+    for t, batch_size in enumerate(batch_size_list):
         start = offset
         end = offset + batch_size
         current_encoder_out = encoder_out.data[start:end]
@@ -2608,7 +2608,6 @@ def modified_beam_search_LODR(
                 context_score = 0
                 new_context_state = None if context_graph is None else hyp.context_state
                 if new_token not in (blank_id, unk_id):
-
                     if context_graph is not None:
                         (
                             context_score,
@@ -2758,7 +2757,7 @@ def modified_beam_search_lm_shallow_fusion(
 
     offset = 0
     finalized_B = []
-    for (t, batch_size) in enumerate(batch_size_list):
+    for t, batch_size in enumerate(batch_size_list):
         start = offset
         end = offset + batch_size
         current_encoder_out = encoder_out.data[start:end]  # get batch
@@ -2900,7 +2899,6 @@ def modified_beam_search_lm_shallow_fusion(
                 new_token = topk_token_indexes[k]
                 new_timestamp = hyp.timestamp[:]
                 if new_token not in (blank_id, unk_id):
-
                     ys.append(new_token)
                     new_timestamp.append(t)
 

egs/librispeech/ASR/pruned_transducer_stateless2/optim.py

@@ -66,7 +66,6 @@ def __init__(
         weight_decay=1e-3,
         target_rms=0.1,
     ):
-
         if not 0.0 <= lr:
             raise ValueError("Invalid learning rate: {}".format(lr))
         if not 0.0 <= eps:

egs/librispeech/ASR/pruned_transducer_stateless2/scaling.py

@@ -528,7 +528,6 @@ def _flatten_parameters(self, flat_weights) -> None:
             return
 
         with torch.cuda.device_of(first_fw):
-
             # Note: no_grad() is necessary since _cudnn_rnn_flatten_weight is
             # an inplace operation on self._flat_weights
             with torch.no_grad():

egs/librispeech/ASR/pruned_transducer_stateless6/vq_utils.py

@@ -56,7 +56,6 @@ class CodebookIndexExtractor:
     """
 
     def __init__(self, params: AttributeDict):
-
         self.params = params
         params.subsets = ["clean-100"]
         if self.params.full_libri:

egs/librispeech/ASR/pruned_transducer_stateless7/alignment.py

@@ -111,7 +111,7 @@ def batch_force_alignment(
 
     offset = 0
     finalized_B = []
-    for (t, batch_size) in enumerate(batch_size_list):
+    for t, batch_size in enumerate(batch_size_list):
         start = offset
         end = offset + batch_size
         current_encoder_out = encoder_out.data[start:end]

egs/librispeech/ASR/streaming_conformer_ctc/train.py

@@ -543,7 +543,6 @@ def train_one_epoch(
             )
 
         if batch_idx % params.log_interval == 0:
-
             if tb_writer is not None:
                 loss_info.write_summary(
                     tb_writer, "train/current_", params.batch_idx_train

egs/librispeech/ASR/tdnn_lstm_ctc/train.py

@@ -463,7 +463,6 @@ def train_one_epoch(
                 f"tot_loss[{tot_loss}], batch size: {batch_size}"
             )
         if batch_idx % params.log_interval == 0:
-
             if tb_writer is not None:
                 loss_info.write_summary(
                     tb_writer, "train/current_", params.batch_idx_train

egs/librispeech/ASR/transducer/train.py

@@ -513,7 +513,6 @@ def train_one_epoch(
             )
 
         if batch_idx % params.log_interval == 0:
-
             if tb_writer is not None:
                 loss_info.write_summary(
                     tb_writer, "train/current_", params.batch_idx_train

egs/librispeech/ASR/transducer_lstm/train.py

@@ -517,7 +517,6 @@ def train_one_epoch(
             )
 
         if batch_idx % params.log_interval == 0:
-
             if tb_writer is not None:
                 loss_info.write_summary(
                     tb_writer, "train/current_", params.batch_idx_train

egs/librispeech/ASR/zipformer/scaling.py

@@ -70,7 +70,7 @@ def __init__(self, *args):
             self.pairs = list(args[0].pairs)
         else:
             self.pairs = [(float(x), float(y)) for x, y in args]
-        for (x, y) in self.pairs:
+        for x, y in self.pairs:
             assert isinstance(x, (float, int)), type(x)
             assert isinstance(y, (float, int)), type(y)
 

icefall/__init__.py

@@ -1,12 +1,6 @@
 # isort:skip_file
 
-from . import (
-    checkpoint,
-    decode,
-    dist,
-    env,
-    utils
-)
+from . import checkpoint, decode, dist, env, utils
 
 from .byte_utils import (
     byte_decode,

icefall/context_graph.py

@@ -227,7 +227,6 @@ def draw(
         filename: Optional[str] = "",
         symbol_table: Optional[Dict[int, str]] = None,
     ) -> "Digraph":  # noqa
-
         """Visualize a ContextGraph via graphviz.
 
         Render ContextGraph as an image via graphviz, and return the Digraph object;