manual correction of black formatting

egs/aidatatang_200zh/ASR/pruned_transducer_stateless2/decode.py

@@ -188,7 +188,7 @@ def get_parser():
         "--context-size",
         type=int,
         default=2,
-        help="The context size in the decoder. 1 means bigram; " "2 means tri-gram",
+        help="The context size in the decoder. 1 means bigram; 2 means tri-gram",
     )
     parser.add_argument(
         "--max-sym-per-frame",

egs/aidatatang_200zh/ASR/pruned_transducer_stateless2/export.py

@@ -103,7 +103,7 @@ def get_parser():
         "--context-size",
         type=int,
         default=2,
-        help="The context size in the decoder. 1 means bigram; " "2 means tri-gram",
+        help="The context size in the decoder. 1 means bigram; 2 means tri-gram",
     )
 
     return parser

egs/aidatatang_200zh/ASR/pruned_transducer_stateless2/pretrained.py

@@ -162,7 +162,7 @@ def get_parser():
         "--context-size",
         type=int,
         default=2,
-        help="The context size in the decoder. 1 means bigram; " "2 means tri-gram",
+        help="The context size in the decoder. 1 means bigram; 2 means tri-gram",
     )
 
     parser.add_argument(
@@ -192,9 +192,9 @@ def read_sound_files(
     ans = []
     for f in filenames:
         wave, sample_rate = torchaudio.load(f)
-        assert sample_rate == expected_sample_rate, (
-            f"expected sample rate: {expected_sample_rate}. " f"Given: {sample_rate}"
-        )
+        assert (
+            sample_rate == expected_sample_rate
+        ), f"expected sample rate: {expected_sample_rate}. Given: {sample_rate}"
         # We use only the first channel
         ans.append(wave[0])
     return ans

egs/aidatatang_200zh/ASR/pruned_transducer_stateless2/train.py

@@ -185,7 +185,7 @@ def get_parser():
         "--context-size",
         type=int,
         default=2,
-        help="The context size in the decoder. 1 means bigram; " "2 means tri-gram",
+        help="The context size in the decoder. 1 means bigram; 2 means tri-gram",
     )
 
     parser.add_argument(
@@ -208,7 +208,7 @@ def get_parser():
         "--am-scale",
         type=float,
         default=0.0,
-        help="The scale to smooth the loss with am (output of encoder network)" "part.",
+        help="The scale to smooth the loss with am (output of encoder network) part.",
     )
 
     parser.add_argument(

egs/aishell/ASR/conformer_ctc/pretrained.py

@@ -210,9 +210,9 @@ def read_sound_files(
     ans = []
     for f in filenames:
         wave, sample_rate = torchaudio.load(f)
-        assert sample_rate == expected_sample_rate, (
-            f"expected sample rate: {expected_sample_rate}. " f"Given: {sample_rate}"
-        )
+        assert (
+            sample_rate == expected_sample_rate
+        ), f"expected sample rate: {expected_sample_rate}. Given: {sample_rate}"
         # We use only the first channel
         ans.append(wave[0])
     return ans

egs/aishell/ASR/pruned_transducer_stateless2/decode.py

@@ -184,7 +184,7 @@ def get_parser():
         "--context-size",
         type=int,
         default=1,
-        help="The context size in the decoder. 1 means bigram; " "2 means tri-gram",
+        help="The context size in the decoder. 1 means bigram; 2 means tri-gram",
     )
     parser.add_argument(
         "--max-sym-per-frame",
@@ -487,7 +487,7 @@ def main():
         ]
         if len(filenames) == 0:
             raise ValueError(
-                f"No checkpoints found for" f" --iter {params.iter}, --avg {params.avg}"
+                f"No checkpoints found for --iter {params.iter}, --avg {params.avg}"
             )
         elif len(filenames) < params.avg:
             raise ValueError(

egs/aishell/ASR/pruned_transducer_stateless2/export.py

@@ -116,7 +116,7 @@ def get_parser():
         "--context-size",
         type=int,
         default=1,
-        help="The context size in the decoder. 1 means bigram; " "2 means tri-gram",
+        help="The context size in the decoder. 1 means bigram; 2 means tri-gram",
     )
 
     add_model_arguments(parser)
@@ -152,7 +152,7 @@ def main():
         ]
         if len(filenames) == 0:
             raise ValueError(
-                f"No checkpoints found for" f" --iter {params.iter}, --avg {params.avg}"
+                f"No checkpoints found for --iter {params.iter}, --avg {params.avg}"
             )
         elif len(filenames) < params.avg:
             raise ValueError(

egs/aishell/ASR/pruned_transducer_stateless2/pretrained.py

@@ -165,7 +165,7 @@ def get_parser():
         "--context-size",
         type=int,
         default=1,
-        help="The context size in the decoder. 1 means bigram; " "2 means tri-gram",
+        help="The context size in the decoder. 1 means bigram; 2 means tri-gram",
     )
     parser.add_argument(
         "--max-sym-per-frame",
@@ -195,9 +195,9 @@ def read_sound_files(
     ans = []
     for f in filenames:
         wave, sample_rate = torchaudio.load(f)
-        assert sample_rate == expected_sample_rate, (
-            f"expected sample rate: {expected_sample_rate}. " f"Given: {sample_rate}"
-        )
+        assert (
+            sample_rate == expected_sample_rate
+        ), f"expected sample rate: {expected_sample_rate}. Given: {sample_rate}"
         # We use only the first channel
         ans.append(wave[0])
     return ans

egs/aishell/ASR/pruned_transducer_stateless2/train.py

@@ -200,7 +200,7 @@ def get_parser():
         "--initial-lr",
         type=float,
         default=0.003,
-        help="The initial learning rate.  This value should not need " "to be changed.",
+        help="The initial learning rate.  This value should not need to be changed.",
     )
 
     parser.add_argument(
@@ -223,7 +223,7 @@ def get_parser():
         "--context-size",
         type=int,
         default=1,
-        help="The context size in the decoder. 1 means bigram; " "2 means tri-gram",
+        help="The context size in the decoder. 1 means bigram; 2 means tri-gram",
     )
 
     parser.add_argument(
@@ -246,7 +246,7 @@ def get_parser():
         "--am-scale",
         type=float,
         default=0.0,
-        help="The scale to smooth the loss with am (output of encoder network)" "part.",
+        help="The scale to smooth the loss with am (output of encoder network) part.",
     )
 
     parser.add_argument(

egs/aishell/ASR/pruned_transducer_stateless3/decode.py

@@ -202,7 +202,7 @@ def get_parser():
         "--context-size",
         type=int,
         default=1,
-        help="The context size in the decoder. 1 means bigram; " "2 means tri-gram",
+        help="The context size in the decoder. 1 means bigram; 2 means tri-gram",
     )
     parser.add_argument(
         "--max-sym-per-frame",

egs/aishell/ASR/pruned_transducer_stateless3/export.py

@@ -132,7 +132,7 @@ def get_parser():
         "--context-size",
         type=int,
         default=1,
-        help="The context size in the decoder. 1 means bigram; " "2 means tri-gram",
+        help="The context size in the decoder. 1 means bigram; 2 means tri-gram",
     )
 
     add_model_arguments(parser)

egs/aishell/ASR/pruned_transducer_stateless3/pretrained.py

@@ -165,7 +165,7 @@ def get_parser():
         "--context-size",
         type=int,
         default=1,
-        help="The context size in the decoder. 1 means bigram; " "2 means tri-gram",
+        help="The context size in the decoder. 1 means bigram; 2 means tri-gram",
     )
     parser.add_argument(
         "--max-sym-per-frame",
@@ -195,9 +195,9 @@ def read_sound_files(
     ans = []
     for f in filenames:
         wave, sample_rate = torchaudio.load(f)
-        assert sample_rate == expected_sample_rate, (
-            f"expected sample rate: {expected_sample_rate}. " f"Given: {sample_rate}"
-        )
+        assert (
+            sample_rate == expected_sample_rate
+        ), f"expected sample rate: {expected_sample_rate}. Given: {sample_rate}"
         # We use only the first channel
         ans.append(wave[0])
     return ans

egs/aishell/ASR/pruned_transducer_stateless3/train.py

@@ -222,7 +222,7 @@ def get_parser():
         "--initial-lr",
         type=float,
         default=0.003,
-        help="The initial learning rate.  This value should not need " "to be changed.",
+        help="The initial learning rate.  This value should not need to be changed.",
     )
 
     parser.add_argument(
@@ -245,7 +245,7 @@ def get_parser():
         "--context-size",
         type=int,
         default=1,
-        help="The context size in the decoder. 1 means bigram; " "2 means tri-gram",
+        help="The context size in the decoder. 1 means bigram; 2 means tri-gram",
     )
 
     parser.add_argument(
@@ -268,7 +268,7 @@ def get_parser():
         "--am-scale",
         type=float,
         default=0.0,
-        help="The scale to smooth the loss with am (output of encoder network)" "part.",
+        help="The scale to smooth the loss with am (output of encoder network) part.",
     )
 
     parser.add_argument(

egs/aishell/ASR/tdnn_lstm_ctc/pretrained.py

@@ -110,9 +110,9 @@ def read_sound_files(
     ans = []
     for f in filenames:
         wave, sample_rate = torchaudio.load(f)
-        assert sample_rate == expected_sample_rate, (
-            f"expected sample rate: {expected_sample_rate}. " f"Given: {sample_rate}"
-        )
+        assert (
+            sample_rate == expected_sample_rate
+        ), f"expected sample rate: {expected_sample_rate}. Given: {sample_rate}"
         # We use only the first channel
         ans.append(wave[0])
     return ans

egs/aishell/ASR/transducer_stateless/decode.py

@@ -99,7 +99,7 @@ def get_parser():
         "--context-size",
         type=int,
         default=2,
-        help="The context size in the decoder. 1 means bigram; " "2 means tri-gram",
+        help="The context size in the decoder. 1 means bigram; 2 means tri-gram",
     )
     parser.add_argument(
         "--max-sym-per-frame",

egs/aishell/ASR/transducer_stateless/export.py

@@ -110,7 +110,7 @@ def get_parser():
         "--context-size",
         type=int,
         default=2,
-        help="The context size in the decoder. 1 means bigram; " "2 means tri-gram",
+        help="The context size in the decoder. 1 means bigram; 2 means tri-gram",
     )
 
     return parser

egs/aishell/ASR/transducer_stateless/pretrained.py

@@ -117,7 +117,7 @@ def get_parser():
         "--context-size",
         type=int,
         default=2,
-        help="The context size in the decoder. 1 means bigram; " "2 means tri-gram",
+        help="The context size in the decoder. 1 means bigram; 2 means tri-gram",
     )
     parser.add_argument(
         "--max-sym-per-frame",
@@ -210,9 +210,9 @@ def read_sound_files(
     ans = []
     for f in filenames:
         wave, sample_rate = torchaudio.load(f)
-        assert sample_rate == expected_sample_rate, (
-            f"expected sample rate: {expected_sample_rate}. " f"Given: {sample_rate}"
-        )
+        assert (
+            sample_rate == expected_sample_rate
+        ), f"expected sample rate: {expected_sample_rate}. Given: {sample_rate}"
         # We use only the first channel
         ans.append(wave[0])
     return ans

egs/aishell/ASR/transducer_stateless/train.py

@@ -126,7 +126,7 @@ def get_parser():
         "--context-size",
         type=int,
         default=2,
-        help="The context size in the decoder. 1 means bigram; " "2 means tri-gram",
+        help="The context size in the decoder. 1 means bigram; 2 means tri-gram",
     )
 
     parser.add_argument(

egs/aishell/ASR/transducer_stateless_modified-2/decode.py

@@ -170,7 +170,7 @@ def get_parser():
         "--context-size",
         type=int,
         default=2,
-        help="The context size in the decoder. 1 means bigram; " "2 means tri-gram",
+        help="The context size in the decoder. 1 means bigram; 2 means tri-gram",
     )
 
     parser.add_argument(

egs/aishell/ASR/transducer_stateless_modified-2/export.py

@@ -109,7 +109,7 @@ def get_parser():
         "--context-size",
         type=int,
         default=2,
-        help="The context size in the decoder. 1 means bigram; " "2 means tri-gram",
+        help="The context size in the decoder. 1 means bigram; 2 means tri-gram",
     )
 
     return parser

egs/aishell/ASR/transducer_stateless_modified-2/pretrained.py

@@ -165,7 +165,7 @@ def get_parser():
         "--context-size",
         type=int,
         default=2,
-        help="The context size in the decoder. 1 means bigram; " "2 means tri-gram",
+        help="The context size in the decoder. 1 means bigram; 2 means tri-gram",
     )
     parser.add_argument(
         "--max-sym-per-frame",
@@ -193,9 +193,9 @@ def read_sound_files(
     ans = []
     for f in filenames:
         wave, sample_rate = torchaudio.load(f)
-        assert sample_rate == expected_sample_rate, (
-            f"expected sample rate: {expected_sample_rate}. " f"Given: {sample_rate}"
-        )
+        assert (
+            sample_rate == expected_sample_rate
+        ), f"expected sample rate: {expected_sample_rate}. Given: {sample_rate}"
         # We use only the first channel
         ans.append(wave[0])
     return ans

egs/aishell/ASR/transducer_stateless_modified-2/train.py

@@ -149,7 +149,7 @@ def get_parser():
         "--context-size",
         type=int,
         default=2,
-        help="The context size in the decoder. 1 means bigram; " "2 means tri-gram",
+        help="The context size in the decoder. 1 means bigram; 2 means tri-gram",
     )
 
     parser.add_argument(
@@ -167,7 +167,7 @@ def get_parser():
         "--datatang-prob",
         type=float,
         default=0.2,
-        help="The probability to select a batch from the " "aidatatang_200zh dataset",
+        help="The probability to select a batch from the aidatatang_200zh dataset",
     )
 
     return parser

egs/aishell/ASR/transducer_stateless_modified/decode.py

@@ -171,7 +171,7 @@ def get_parser():
         "--context-size",
         type=int,
         default=2,
-        help="The context size in the decoder. 1 means bigram; " "2 means tri-gram",
+        help="The context size in the decoder. 1 means bigram; 2 means tri-gram",
     )
 
     parser.add_argument(

egs/aishell/ASR/transducer_stateless_modified/export.py

@@ -109,7 +109,7 @@ def get_parser():
         "--context-size",
         type=int,
         default=2,
-        help="The context size in the decoder. 1 means bigram; " "2 means tri-gram",
+        help="The context size in the decoder. 1 means bigram; 2 means tri-gram",
     )
 
     return parser

egs/aishell/ASR/transducer_stateless_modified/pretrained.py

@@ -165,7 +165,7 @@ def get_parser():
         "--context-size",
         type=int,
         default=2,
-        help="The context size in the decoder. 1 means bigram; " "2 means tri-gram",
+        help="The context size in the decoder. 1 means bigram; 2 means tri-gram",
     )
     parser.add_argument(
         "--max-sym-per-frame",
@@ -193,9 +193,9 @@ def read_sound_files(
     ans = []
     for f in filenames:
         wave, sample_rate = torchaudio.load(f)
-        assert sample_rate == expected_sample_rate, (
-            f"expected sample rate: {expected_sample_rate}. " f"Given: {sample_rate}"
-        )
+        assert (
+            sample_rate == expected_sample_rate
+        ), f"expected sample rate: {expected_sample_rate}. Given: {sample_rate}"
         # We use only the first channel
         ans.append(wave[0])
     return ans

egs/aishell/ASR/transducer_stateless_modified/train.py

@@ -142,7 +142,7 @@ def get_parser():
         "--context-size",
         type=int,
         default=2,
-        help="The context size in the decoder. 1 means bigram; " "2 means tri-gram",
+        help="The context size in the decoder. 1 means bigram; 2 means tri-gram",
     )
 
     parser.add_argument(

egs/aishell2/ASR/pruned_transducer_stateless5/decode.py

@@ -269,7 +269,7 @@ def get_parser():
         "--context-size",
         type=int,
         default=2,
-        help="The context size in the decoder. 1 means bigram; " "2 means tri-gram",
+        help="The context size in the decoder. 1 means bigram; 2 means tri-gram",
     )
     parser.add_argument(
         "--max-sym-per-frame",

egs/aishell2/ASR/pruned_transducer_stateless5/export.py

@@ -133,7 +133,7 @@ def get_parser():
         "--context-size",
         type=int,
         default=2,
-        help="The context size in the decoder. 1 means bigram; " "2 means tri-gram",
+        help="The context size in the decoder. 1 means bigram; 2 means tri-gram",
     )
 
     add_model_arguments(parser)

egs/aishell2/ASR/pruned_transducer_stateless5/pretrained.py

@@ -159,7 +159,7 @@ def get_parser():
         "--context-size",
         type=int,
         default=2,
-        help="The context size in the decoder. 1 means bigram; " "2 means tri-gram",
+        help="The context size in the decoder. 1 means bigram; 2 means tri-gram",
     )
     parser.add_argument(
         "--max-sym-per-frame",
@@ -190,9 +190,9 @@ def read_sound_files(
     ans = []
     for f in filenames:
         wave, sample_rate = torchaudio.load(f)
-        assert sample_rate == expected_sample_rate, (
-            f"expected sample rate: {expected_sample_rate}. " f"Given: {sample_rate}"
-        )
+        assert (
+            sample_rate == expected_sample_rate
+        ), f"expected sample rate: {expected_sample_rate}. Given: {sample_rate}"
         # We use only the first channel
         ans.append(wave[0])
     return ans

egs/aishell2/ASR/pruned_transducer_stateless5/train.py

@@ -218,7 +218,7 @@ def get_parser():
         "--initial-lr",
         type=float,
         default=0.003,
-        help="The initial learning rate.  This value should not need " "to be changed.",
+        help="The initial learning rate.  This value should not need to be changed.",
     )
 
     parser.add_argument(
@@ -241,7 +241,7 @@ def get_parser():
         "--context-size",
         type=int,
         default=2,
-        help="The context size in the decoder. 1 means bigram; " "2 means tri-gram",
+        help="The context size in the decoder. 1 means bigram; 2 means tri-gram",
     )
 
     parser.add_argument(
@@ -264,7 +264,7 @@ def get_parser():
         "--am-scale",
         type=float,
         default=0.0,
-        help="The scale to smooth the loss with am (output of encoder network)" "part.",
+        help="The scale to smooth the loss with am (output of encoder network) part.",
     )
 
     parser.add_argument(

egs/aishell4/ASR/pruned_transducer_stateless5/decode.py

@@ -201,7 +201,7 @@ def get_parser():
         "--context-size",
         type=int,
         default=2,
-        help="The context size in the decoder. 1 means bigram; " "2 means tri-gram",
+        help="The context size in the decoder. 1 means bigram; 2 means tri-gram",
     )
     parser.add_argument(
         "--max-sym-per-frame",

egs/aishell4/ASR/pruned_transducer_stateless5/export.py

@@ -136,7 +136,7 @@ def get_parser():
         "--context-size",
         type=int,
         default=2,
-        help="The context size in the decoder. 1 means bigram; " "2 means tri-gram",
+        help="The context size in the decoder. 1 means bigram; 2 means tri-gram",
     )
 
     add_model_arguments(parser)

egs/aishell4/ASR/pruned_transducer_stateless5/pretrained.py

@@ -172,7 +172,7 @@ def get_parser():
         "--context-size",
         type=int,
         default=2,
-        help="The context size in the decoder. 1 means bigram; " "2 means tri-gram",
+        help="The context size in the decoder. 1 means bigram; 2 means tri-gram",
     )
     parser.add_argument(
         "--max-sym-per-frame",
@@ -203,9 +203,9 @@ def read_sound_files(
     ans = []
     for f in filenames:
         wave, sample_rate = torchaudio.load(f)
-        assert sample_rate == expected_sample_rate, (
-            f"expected sample rate: {expected_sample_rate}. " f"Given: {sample_rate}"
-        )
+        assert (
+            sample_rate == expected_sample_rate
+        ), f"expected sample rate: {expected_sample_rate}. Given: {sample_rate}"
         # We use only the first channel
         ans.append(wave[0])
     return ans

egs/aishell4/ASR/pruned_transducer_stateless5/train.py

@@ -211,7 +211,7 @@ def get_parser():
         "--initial-lr",
         type=float,
         default=0.003,
-        help="The initial learning rate.  This value should not need " "to be changed.",
+        help="The initial learning rate.  This value should not need to be changed.",
     )
 
     parser.add_argument(
@@ -234,7 +234,7 @@ def get_parser():
         "--context-size",
         type=int,
         default=2,
-        help="The context size in the decoder. 1 means bigram; " "2 means tri-gram",
+        help="The context size in the decoder. 1 means bigram; 2 means tri-gram",
     )
 
     parser.add_argument(
@@ -257,7 +257,7 @@ def get_parser():
         "--am-scale",
         type=float,
         default=0.0,
-        help="The scale to smooth the loss with am (output of encoder network)" "part.",
+        help="The scale to smooth the loss with am (output of encoder network) part.",
     )
 
     parser.add_argument(

egs/alimeeting/ASR/pruned_transducer_stateless2/decode.py

@@ -189,7 +189,7 @@ def get_parser():
         "--context-size",
         type=int,
         default=2,
-        help="The context size in the decoder. 1 means bigram; " "2 means tri-gram",
+        help="The context size in the decoder. 1 means bigram; 2 means tri-gram",
     )
     parser.add_argument(
         "--max-sym-per-frame",

egs/alimeeting/ASR/pruned_transducer_stateless2/export.py

@@ -103,7 +103,7 @@ def get_parser():
         "--context-size",
         type=int,
         default=2,
-        help="The context size in the decoder. 1 means bigram; " "2 means tri-gram",
+        help="The context size in the decoder. 1 means bigram; 2 means tri-gram",
     )
 
     return parser

egs/alimeeting/ASR/pruned_transducer_stateless2/pretrained.py

@@ -162,7 +162,7 @@ def get_parser():
         "--context-size",
         type=int,
         default=2,
-        help="The context size in the decoder. 1 means bigram; " "2 means tri-gram",
+        help="The context size in the decoder. 1 means bigram; 2 means tri-gram",
     )
 
     parser.add_argument(
@@ -192,9 +192,9 @@ def read_sound_files(
     ans = []
     for f in filenames:
         wave, sample_rate = torchaudio.load(f)
-        assert sample_rate == expected_sample_rate, (
-            f"expected sample rate: {expected_sample_rate}. " f"Given: {sample_rate}"
-        )
+        assert (
+            sample_rate == expected_sample_rate
+        ), f"expected sample rate: {expected_sample_rate}. Given: {sample_rate}"
         # We use only the first channel
         ans.append(wave[0])
     return ans

egs/alimeeting/ASR/pruned_transducer_stateless2/train.py

@@ -185,7 +185,7 @@ def get_parser():
         "--context-size",
         type=int,
         default=2,
-        help="The context size in the decoder. 1 means bigram; " "2 means tri-gram",
+        help="The context size in the decoder. 1 means bigram; 2 means tri-gram",
     )
 
     parser.add_argument(
@@ -208,7 +208,7 @@ def get_parser():
         "--am-scale",
         type=float,
         default=0.0,
-        help="The scale to smooth the loss with am (output of encoder network)" "part.",
+        help="The scale to smooth the loss with am (output of encoder network) part.",
     )
 
     parser.add_argument(

egs/csj/ASR/local/compute_fbank_csj.py

@@ -25,7 +25,9 @@ from random import Random
 from typing import List, Tuple
 
 import torch
-from lhotse import (  # fmt: off; See the following for why LilcomChunkyWriter is preferred; https://github.com/k2-fsa/icefall/pull/404; https://github.com/lhotse-speech/lhotse/pull/527; fmt: on
+
+# fmt: off
+from lhotse import (  # See the following for why LilcomChunkyWriter is preferred; https://github.com/k2-fsa/icefall/pull/404; https://github.com/lhotse-speech/lhotse/pull/527
     CutSet,
     Fbank,
     FbankConfig,
@@ -34,6 +36,8 @@ from lhotse import (  # fmt: off; See the following for why LilcomChunkyWriter i
     SupervisionSet,
 )
 
+# fmt: on
+
 ARGPARSE_DESCRIPTION = """
 This script follows the espnet method of splitting the remaining core+noncore
 utterances into valid and train cutsets at an index which is by default 4000.
@@ -92,7 +96,7 @@ def make_cutset_blueprints(
     cut_set = cut_set.shuffle(Random(RNG_SEED))
 
     logging.info(
-        "Creating valid and train cuts from core and noncore," f"split at {split}."
+        "Creating valid and train cuts from core and noncore, split at {split}."
     )
     valid_set = CutSet.from_cuts(islice(cut_set, 0, split))
 

egs/csj/ASR/local/prepare_lang_char.py

@@ -87,7 +87,7 @@ def main():
     args = get_args()
 
     logging.basicConfig(
-        format=("%(asctime)s %(levelname)s [%(filename)s:%(lineno)d] " "%(message)s"),
+        format=("%(asctime)s %(levelname)s [%(filename)s:%(lineno)d] %(message)s"),
         level=logging.INFO,
     )
 
@@ -109,7 +109,7 @@ def main():
 
     words = set()
     logging.info(
-        f"Creating vocabulary from {args.train_cut.name}" f" at {args.trans_mode} mode."
+        f"Creating vocabulary from {args.train_cut.name} at {args.trans_mode} mode."
     )
     for cut in train_set:
         try:
@@ -120,7 +120,7 @@ def main():
             )
         except KeyError:
             raise KeyError(
-                f"Could not find {args.trans_mode} in " f"{cut.supervisions[0].custom}"
+                f"Could not find {args.trans_mode} in {cut.supervisions[0].custom}"
             )
         for t in text.split():
             if t in args.userdef_string:

egs/gigaspeech/ASR/conformer_ctc/asr_datamodule.py

@@ -183,7 +183,7 @@ class GigaSpeechAsrDataModule:
             "--small-dev",
             type=str2bool,
             default=False,
-            help="Should we use only 1000 utterances for dev " "(speeds up training)",
+            help="Should we use only 1000 utterances for dev (speeds up training)",
         )
 
     def train_dataloaders(self, cuts_train: CutSet) -> DataLoader:

egs/gigaspeech/ASR/pruned_transducer_stateless2/asr_datamodule.py

@@ -195,7 +195,7 @@ class GigaSpeechAsrDataModule:
             "--small-dev",
             type=str2bool,
             default=False,
-            help="Should we use only 1000 utterances for dev " "(speeds up training)",
+            help="Should we use only 1000 utterances for dev (speeds up training)",
         )
 
     def train_dataloaders(

egs/gigaspeech/ASR/pruned_transducer_stateless2/decode.py

@@ -184,7 +184,7 @@ def get_parser():
         "--context-size",
         type=int,
         default=2,
-        help="The context size in the decoder. 1 means bigram; " "2 means tri-gram",
+        help="The context size in the decoder. 1 means bigram; 2 means tri-gram",
     )
     parser.add_argument(
         "--max-sym-per-frame",
@@ -498,7 +498,7 @@ def main():
         ]
         if len(filenames) == 0:
             raise ValueError(
-                f"No checkpoints found for" f" --iter {params.iter}, --avg {params.avg}"
+                f"No checkpoints found for --iter {params.iter}, --avg {params.avg}"
             )
         elif len(filenames) < params.avg:
             raise ValueError(

egs/gigaspeech/ASR/pruned_transducer_stateless2/export.py

@@ -116,7 +116,7 @@ def get_parser():
         "--context-size",
         type=int,
         default=2,
-        help="The context size in the decoder. 1 means bigram; " "2 means tri-gram",
+        help="The context size in the decoder. 1 means bigram; 2 means tri-gram",
     )
 
     return parser
@@ -155,7 +155,7 @@ def main():
         ]
         if len(filenames) == 0:
             raise ValueError(
-                f"No checkpoints found for" f" --iter {params.iter}, --avg {params.avg}"
+                f"No checkpoints found for --iter {params.iter}, --avg {params.avg}"
             )
         elif len(filenames) < params.avg:
             raise ValueError(

egs/gigaspeech/ASR/pruned_transducer_stateless2/train.py

@@ -176,7 +176,7 @@ def get_parser():
         "--context-size",
         type=int,
         default=2,
-        help="The context size in the decoder. 1 means bigram; " "2 means tri-gram",
+        help="The context size in the decoder. 1 means bigram; 2 means tri-gram",
     )
 
     parser.add_argument(
@@ -199,7 +199,7 @@ def get_parser():
         "--am-scale",
         type=float,
         default=0.0,
-        help="The scale to smooth the loss with am (output of encoder network)" "part.",
+        help="The scale to smooth the loss with am (output of encoder network) part.",
     )
 
     parser.add_argument(

egs/librispeech/ASR/conformer_ctc/pretrained.py

@@ -236,9 +236,9 @@ def read_sound_files(
     ans = []
     for f in filenames:
         wave, sample_rate = torchaudio.load(f)
-        assert sample_rate == expected_sample_rate, (
-            f"expected sample rate: {expected_sample_rate}. " f"Given: {sample_rate}"
-        )
+        assert (
+            sample_rate == expected_sample_rate
+        ), f"expected sample rate: {expected_sample_rate}. Given: {sample_rate}"
         # We use only the first channel
         ans.append(wave[0])
     return ans

egs/librispeech/ASR/conv_emformer_transducer_stateless/decode.py

@@ -215,7 +215,7 @@ def get_parser():
         "--context-size",
         type=int,
         default=2,
-        help="The context size in the decoder. 1 means bigram; " "2 means tri-gram",
+        help="The context size in the decoder. 1 means bigram; 2 means tri-gram",
     )
 
     parser.add_argument(

egs/librispeech/ASR/conv_emformer_transducer_stateless/emformer.py

@@ -445,7 +445,7 @@ class EmformerAttention(nn.Module):
 
         if embed_dim % nhead != 0:
             raise ValueError(
-                f"embed_dim ({embed_dim}) is not a multiple of" f"nhead ({nhead})."
+                f"embed_dim ({embed_dim}) is not a multiple of nhead ({nhead})."
             )
 
         self.embed_dim = embed_dim

egs/librispeech/ASR/conv_emformer_transducer_stateless/export.py

@@ -136,7 +136,7 @@ def get_parser():
         "--context-size",
         type=int,
         default=2,
-        help="The context size in the decoder. 1 means bigram; " "2 means tri-gram",
+        help="The context size in the decoder. 1 means bigram; 2 means tri-gram",
     )
 
     parser.add_argument(

egs/librispeech/ASR/conv_emformer_transducer_stateless/streaming_decode.py

@@ -211,7 +211,7 @@ def get_parser():
         "--context-size",
         type=int,
         default=2,
-        help="The context size in the decoder. 1 means bigram; " "2 means tri-gram",
+        help="The context size in the decoder. 1 means bigram; 2 means tri-gram",
     )
     parser.add_argument(
         "--max-sym-per-frame",

egs/librispeech/ASR/conv_emformer_transducer_stateless/train.py

@@ -263,7 +263,7 @@ def get_parser():
         "--context-size",
         type=int,
         default=2,
-        help="The context size in the decoder. 1 means bigram; " "2 means tri-gram",
+        help="The context size in the decoder. 1 means bigram; 2 means tri-gram",
     )
 
     parser.add_argument(
@@ -286,7 +286,7 @@ def get_parser():
         "--am-scale",
         type=float,
         default=0.0,
-        help="The scale to smooth the loss with am (output of encoder network)" "part.",
+        help="The scale to smooth the loss with am (output of encoder network) part.",
     )
 
     parser.add_argument(

egs/librispeech/ASR/conv_emformer_transducer_stateless2/decode.py

@@ -215,7 +215,7 @@ def get_parser():
         "--context-size",
         type=int,
         default=2,
-        help="The context size in the decoder. 1 means bigram; " "2 means tri-gram",
+        help="The context size in the decoder. 1 means bigram; 2 means tri-gram",
     )
 
     parser.add_argument(

egs/librispeech/ASR/conv_emformer_transducer_stateless2/emformer.py

@@ -445,7 +445,7 @@ class EmformerAttention(nn.Module):
 
         if embed_dim % nhead != 0:
             raise ValueError(
-                f"embed_dim ({embed_dim}) is not a multiple of" f"nhead ({nhead})."
+                f"embed_dim ({embed_dim}) is not a multiple of nhead ({nhead})."
             )
 
         self.embed_dim = embed_dim

egs/librispeech/ASR/conv_emformer_transducer_stateless2/export.py

@@ -136,7 +136,7 @@ def get_parser():
         "--context-size",
         type=int,
         default=2,
-        help="The context size in the decoder. 1 means bigram; " "2 means tri-gram",
+        help="The context size in the decoder. 1 means bigram; 2 means tri-gram",
     )
 
     parser.add_argument(

egs/librispeech/ASR/conv_emformer_transducer_stateless2/streaming_decode.py

@@ -211,7 +211,7 @@ def get_parser():
         "--context-size",
         type=int,
         default=2,
-        help="The context size in the decoder. 1 means bigram; " "2 means tri-gram",
+        help="The context size in the decoder. 1 means bigram; 2 means tri-gram",
     )
     parser.add_argument(
         "--max-sym-per-frame",

egs/librispeech/ASR/conv_emformer_transducer_stateless2/train.py

@@ -263,7 +263,7 @@ def get_parser():
         "--context-size",
         type=int,
         default=2,
-        help="The context size in the decoder. 1 means bigram; " "2 means tri-gram",
+        help="The context size in the decoder. 1 means bigram; 2 means tri-gram",
     )
 
     parser.add_argument(
@@ -286,7 +286,7 @@ def get_parser():
         "--am-scale",
         type=float,
         default=0.0,
-        help="The scale to smooth the loss with am (output of encoder network)" "part.",
+        help="The scale to smooth the loss with am (output of encoder network) part.",
     )
 
     parser.add_argument(

egs/librispeech/ASR/local/filter_cuts.py

@@ -79,7 +79,7 @@ def filter_cuts(cut_set: CutSet, sp: spm.SentencePieceProcessor):
         total += 1
         if c.duration < 1.0 or c.duration > 20.0:
             logging.warning(
-                f"Exclude cut with ID {c.id} from training. " f"Duration: {c.duration}"
+                f"Exclude cut with ID {c.id} from training. Duration: {c.duration}"
             )
             removed += 1
             return False
@@ -124,7 +124,7 @@ def filter_cuts(cut_set: CutSet, sp: spm.SentencePieceProcessor):
     ans = cut_set.filter(remove_short_and_long_utterances).to_eager()
     ratio = removed / total * 100
     logging.info(
-        f"Removed {removed} cuts from {total} cuts. " f"{ratio:.3f}% data is removed."
+        f"Removed {removed} cuts from {total} cuts. {ratio:.3f}% data is removed."
     )
     return ans
 

egs/librispeech/ASR/local/prepare_lm_training_data.py

@@ -137,7 +137,7 @@ def main():
     for i in range(num_sentences):
         if step and i % step == 0:
             logging.info(
-                f"Processed number of lines: {i} " f"({i/num_sentences*100: .3f}%)"
+                f"Processed number of lines: {i} ({i/num_sentences*100: .3f}%)"
             )
 
         word_ids = sentences[i]

egs/librispeech/ASR/lstm_transducer_stateless/decode.py

@@ -272,7 +272,7 @@ def get_parser():
         "--context-size",
         type=int,
         default=2,
-        help="The context size in the decoder. 1 means bigram; " "2 means tri-gram",
+        help="The context size in the decoder. 1 means bigram; 2 means tri-gram",
     )
 
     parser.add_argument(

egs/librispeech/ASR/lstm_transducer_stateless/export.py

@@ -172,7 +172,7 @@ def get_parser():
         "--context-size",
         type=int,
         default=2,
-        help="The context size in the decoder. 1 means bigram; " "2 means tri-gram",
+        help="The context size in the decoder. 1 means bigram; 2 means tri-gram",
     )
 
     add_model_arguments(parser)

egs/librispeech/ASR/lstm_transducer_stateless/jit_pretrained.py

@@ -123,9 +123,9 @@ def read_sound_files(
     ans = []
     for f in filenames:
         wave, sample_rate = torchaudio.load(f)
-        assert sample_rate == expected_sample_rate, (
-            f"expected sample rate: {expected_sample_rate}. " f"Given: {sample_rate}"
-        )
+        assert (
+            sample_rate == expected_sample_rate
+        ), f"expected sample rate: {expected_sample_rate}. Given: {sample_rate}"
         # We use only the first channel
         ans.append(wave[0])
     return ans

egs/librispeech/ASR/lstm_transducer_stateless/pretrained.py

@@ -166,7 +166,7 @@ def get_parser():
         "--context-size",
         type=int,
         default=2,
-        help="The context size in the decoder. 1 means bigram; " "2 means tri-gram",
+        help="The context size in the decoder. 1 means bigram; 2 means tri-gram",
     )
     parser.add_argument(
         "--max-sym-per-frame",
@@ -197,9 +197,9 @@ def read_sound_files(
     ans = []
     for f in filenames:
         wave, sample_rate = torchaudio.load(f)
-        assert sample_rate == expected_sample_rate, (
-            f"expected sample rate: {expected_sample_rate}. " f"Given: {sample_rate}"
-        )
+        assert (
+            sample_rate == expected_sample_rate
+        ), f"expected sample rate: {expected_sample_rate}. Given: {sample_rate}"
         # We use only the first channel
         ans.append(wave[0])
     return ans

egs/librispeech/ASR/lstm_transducer_stateless/streaming_decode.py

@@ -199,7 +199,7 @@ def get_parser():
         "--context-size",
         type=int,
         default=2,
-        help="The context size in the decoder. 1 means bigram; " "2 means tri-gram",
+        help="The context size in the decoder. 1 means bigram; 2 means tri-gram",
     )
     parser.add_argument(
         "--max-sym-per-frame",

egs/librispeech/ASR/lstm_transducer_stateless/train.py

@@ -220,7 +220,7 @@ def get_parser():
         "--context-size",
         type=int,
         default=2,
-        help="The context size in the decoder. 1 means bigram; " "2 means tri-gram",
+        help="The context size in the decoder. 1 means bigram; 2 means tri-gram",
     )
 
     parser.add_argument(
@@ -243,7 +243,7 @@ def get_parser():
         "--am-scale",
         type=float,
         default=0.0,
-        help="The scale to smooth the loss with am (output of encoder network)" "part.",
+        help="The scale to smooth the loss with am (output of encoder network) part.",
     )
 
     parser.add_argument(
@@ -970,7 +970,7 @@ def run(rank, world_size, args):
         # the threshold
         if c.duration < 1.0 or c.duration > 20.0:
             logging.warning(
-                f"Exclude cut with ID {c.id} from training. " f"Duration: {c.duration}"
+                f"Exclude cut with ID {c.id} from training. Duration: {c.duration}"
             )
             return False
 

egs/librispeech/ASR/lstm_transducer_stateless2/decode.py

@@ -295,7 +295,7 @@ def get_parser():
         "--context-size",
         type=int,
         default=2,
-        help="The context size in the decoder. 1 means bigram; " "2 means tri-gram",
+        help="The context size in the decoder. 1 means bigram; 2 means tri-gram",
     )
 
     parser.add_argument(

egs/librispeech/ASR/lstm_transducer_stateless2/export.py

@@ -225,7 +225,7 @@ def get_parser():
         "--context-size",
         type=int,
         default=2,
-        help="The context size in the decoder. 1 means bigram; " "2 means tri-gram",
+        help="The context size in the decoder. 1 means bigram; 2 means tri-gram",
     )
 
     add_model_arguments(parser)

egs/librispeech/ASR/lstm_transducer_stateless2/jit_pretrained.py

@@ -124,9 +124,9 @@ def read_sound_files(
     ans = []
     for f in filenames:
         wave, sample_rate = torchaudio.load(f)
-        assert sample_rate == expected_sample_rate, (
-            f"expected sample rate: {expected_sample_rate}. " f"Given: {sample_rate}"
-        )
+        assert (
+            sample_rate == expected_sample_rate
+        ), f"expected sample rate: {expected_sample_rate}. Given: {sample_rate}"
         # We use only the first channel
         ans.append(wave[0])
     return ans

egs/librispeech/ASR/lstm_transducer_stateless2/ncnn-decode.py

@@ -198,9 +198,9 @@ def read_sound_files(
     ans = []
     for f in filenames:
         wave, sample_rate = torchaudio.load(f)
-        assert sample_rate == expected_sample_rate, (
-            f"expected sample rate: {expected_sample_rate}. " f"Given: {sample_rate}"
-        )
+        assert (
+            sample_rate == expected_sample_rate
+        ), f"expected sample rate: {expected_sample_rate}. Given: {sample_rate}"
         # We use only the first channel
         ans.append(wave[0])
     return ans

egs/librispeech/ASR/lstm_transducer_stateless2/pretrained.py

@@ -169,7 +169,7 @@ def get_parser():
         "--context-size",
         type=int,
         default=2,
-        help="The context size in the decoder. 1 means bigram; " "2 means tri-gram",
+        help="The context size in the decoder. 1 means bigram; 2 means tri-gram",
     )
     parser.add_argument(
         "--max-sym-per-frame",
@@ -200,9 +200,9 @@ def read_sound_files(
     ans = []
     for f in filenames:
         wave, sample_rate = torchaudio.load(f)
-        assert sample_rate == expected_sample_rate, (
-            f"expected sample rate: {expected_sample_rate}. " f"Given: {sample_rate}"
-        )
+        assert (
+            sample_rate == expected_sample_rate
+        ), f"expected sample rate: {expected_sample_rate}. Given: {sample_rate}"
         # We use only the first channel
         ans.append(wave[0])
     return ans

egs/librispeech/ASR/lstm_transducer_stateless2/streaming-ncnn-decode.py

@@ -186,9 +186,9 @@ def read_sound_files(
     ans = []
     for f in filenames:
         wave, sample_rate = torchaudio.load(f)
-        assert sample_rate == expected_sample_rate, (
-            f"expected sample rate: {expected_sample_rate}. " f"Given: {sample_rate}"
-        )
+        assert (
+            sample_rate == expected_sample_rate
+        ), f"expected sample rate: {expected_sample_rate}. Given: {sample_rate}"
         # We use only the first channel
         ans.append(wave[0])
     return ans

egs/librispeech/ASR/lstm_transducer_stateless2/streaming-onnx-decode.py

@@ -147,9 +147,9 @@ def read_sound_files(
     ans = []
     for f in filenames:
         wave, sample_rate = torchaudio.load(f)
-        assert sample_rate == expected_sample_rate, (
-            f"expected sample rate: {expected_sample_rate}. " f"Given: {sample_rate}"
-        )
+        assert (
+            sample_rate == expected_sample_rate
+        ), f"expected sample rate: {expected_sample_rate}. Given: {sample_rate}"
         # We use only the first channel
         ans.append(wave[0])
     return ans

egs/librispeech/ASR/lstm_transducer_stateless2/train.py

@@ -161,7 +161,7 @@ def get_parser():
         "--full-libri",
         type=str2bool,
         default=True,
-        help="When enabled, use 960h LibriSpeech. " "Otherwise, use 100h subset.",
+        help="When enabled, use 960h LibriSpeech. Otherwise, use 100h subset.",
     )
 
     parser.add_argument(
@@ -235,7 +235,7 @@ def get_parser():
         "--context-size",
         type=int,
         default=2,
-        help="The context size in the decoder. 1 means bigram; " "2 means tri-gram",
+        help="The context size in the decoder. 1 means bigram; 2 means tri-gram",
     )
 
     parser.add_argument(
@@ -258,7 +258,7 @@ def get_parser():
         "--am-scale",
         type=float,
         default=0.0,
-        help="The scale to smooth the loss with am (output of encoder network)" "part.",
+        help="The scale to smooth the loss with am (output of encoder network) part.",
     )
 
     parser.add_argument(
@@ -986,7 +986,7 @@ def filter_short_and_long_utterances(
         # the threshold
         if c.duration < 1.0 or c.duration > 20.0:
             logging.warning(
-                f"Exclude cut with ID {c.id} from training. " f"Duration: {c.duration}"
+                f"Exclude cut with ID {c.id} from training. Duration: {c.duration}"
             )
             return False
 

egs/librispeech/ASR/lstm_transducer_stateless3/decode.py

@@ -290,7 +290,7 @@ def get_parser():
         "--context-size",
         type=int,
         default=2,
-        help="The context size in the decoder. 1 means bigram; " "2 means tri-gram",
+        help="The context size in the decoder. 1 means bigram; 2 means tri-gram",
     )
 
     parser.add_argument(

egs/librispeech/ASR/lstm_transducer_stateless3/export.py

@@ -172,7 +172,7 @@ def get_parser():
         "--context-size",
         type=int,
         default=2,
-        help="The context size in the decoder. 1 means bigram; " "2 means tri-gram",
+        help="The context size in the decoder. 1 means bigram; 2 means tri-gram",
     )
 
     add_model_arguments(parser)

egs/librispeech/ASR/lstm_transducer_stateless3/jit_pretrained.py

@@ -123,9 +123,9 @@ def read_sound_files(
     ans = []
     for f in filenames:
         wave, sample_rate = torchaudio.load(f)
-        assert sample_rate == expected_sample_rate, (
-            f"expected sample rate: {expected_sample_rate}. " f"Given: {sample_rate}"
-        )
+        assert (
+            sample_rate == expected_sample_rate
+        ), f"expected sample rate: {expected_sample_rate}. Given: {sample_rate}"
         # We use only the first channel
         ans.append(wave[0])
     return ans

egs/librispeech/ASR/lstm_transducer_stateless3/pretrained.py

@@ -166,7 +166,7 @@ def get_parser():
         "--context-size",
         type=int,
         default=2,
-        help="The context size in the decoder. 1 means bigram; " "2 means tri-gram",
+        help="The context size in the decoder. 1 means bigram; 2 means tri-gram",
     )
     parser.add_argument(
         "--max-sym-per-frame",
@@ -197,9 +197,9 @@ def read_sound_files(
     ans = []
     for f in filenames:
         wave, sample_rate = torchaudio.load(f)
-        assert sample_rate == expected_sample_rate, (
-            f"expected sample rate: {expected_sample_rate}. " f"Given: {sample_rate}"
-        )
+        assert (
+            sample_rate == expected_sample_rate
+        ), f"expected sample rate: {expected_sample_rate}. Given: {sample_rate}"
         # We use only the first channel
         ans.append(wave[0])
     return ans

egs/librispeech/ASR/lstm_transducer_stateless3/streaming_decode.py

@@ -199,7 +199,7 @@ def get_parser():
         "--context-size",
         type=int,
         default=2,
-        help="The context size in the decoder. 1 means bigram; " "2 means tri-gram",
+        help="The context size in the decoder. 1 means bigram; 2 means tri-gram",
     )
     parser.add_argument(
         "--max-sym-per-frame",

egs/librispeech/ASR/lstm_transducer_stateless3/train.py

@@ -230,7 +230,7 @@ def get_parser():
         "--context-size",
         type=int,
         default=2,
-        help="The context size in the decoder. 1 means bigram; " "2 means tri-gram",
+        help="The context size in the decoder. 1 means bigram; 2 means tri-gram",
     )
 
     parser.add_argument(
@@ -253,7 +253,7 @@ def get_parser():
         "--am-scale",
         type=float,
         default=0.0,
-        help="The scale to smooth the loss with am (output of encoder network)" "part.",
+        help="The scale to smooth the loss with am (output of encoder network) part.",
     )
 
     parser.add_argument(
@@ -987,7 +987,7 @@ def run(rank, world_size, args):
         # the threshold
         if c.duration < 1.0 or c.duration > 20.0:
             logging.warning(
-                f"Exclude cut with ID {c.id} from training. " f"Duration: {c.duration}"
+                f"Exclude cut with ID {c.id} from training. Duration: {c.duration}"
             )
             return False
 

egs/librispeech/ASR/pruned2_knowledge/asr_datamodule.py

@@ -83,7 +83,7 @@ class LibriSpeechAsrDataModule:
             "--full-libri",
             type=str2bool,
             default=True,
-            help="When enabled, use 960h LibriSpeech. " "Otherwise, use 100h subset.",
+            help="When enabled, use 960h LibriSpeech. Otherwise, use 100h subset.",
         )
         group.add_argument(
             "--manifest-dir",

egs/librispeech/ASR/pruned2_knowledge/decode.py

@@ -182,7 +182,7 @@ def get_parser():
         "--context-size",
         type=int,
         default=2,
-        help="The context size in the decoder. 1 means bigram; " "2 means tri-gram",
+        help="The context size in the decoder. 1 means bigram; 2 means tri-gram",
     )
     parser.add_argument(
         "--max-sym-per-frame",

egs/librispeech/ASR/pruned2_knowledge/export.py

@@ -105,7 +105,7 @@ def get_parser():
         "--context-size",
         type=int,
         default=2,
-        help="The context size in the decoder. 1 means bigram; " "2 means tri-gram",
+        help="The context size in the decoder. 1 means bigram; 2 means tri-gram",
     )
 
     return parser

egs/librispeech/ASR/pruned2_knowledge/train.py

@@ -177,7 +177,7 @@ def get_parser():
         "--context-size",
         type=int,
         default=2,
-        help="The context size in the decoder. 1 means bigram; " "2 means tri-gram",
+        help="The context size in the decoder. 1 means bigram; 2 means tri-gram",
     )
 
     parser.add_argument(
@@ -200,7 +200,7 @@ def get_parser():
         "--am-scale",
         type=float,
         default=0.0,
-        help="The scale to smooth the loss with am (output of encoder network)" "part.",
+        help="The scale to smooth the loss with am (output of encoder network) part.",
     )
 
     parser.add_argument(

egs/librispeech/ASR/pruned_stateless_emformer_rnnt2/decode.py

@@ -204,7 +204,7 @@ def get_parser():
         "--context-size",
         type=int,
         default=2,
-        help="The context size in the decoder. 1 means bigram; " "2 means tri-gram",
+        help="The context size in the decoder. 1 means bigram; 2 means tri-gram",
     )
     parser.add_argument(
         "--max-sym-per-frame",

egs/librispeech/ASR/pruned_stateless_emformer_rnnt2/export.py

@@ -133,7 +133,7 @@ def get_parser():
         "--context-size",
         type=int,
         default=2,
-        help="The context size in the decoder. 1 means bigram; " "2 means tri-gram",
+        help="The context size in the decoder. 1 means bigram; 2 means tri-gram",
     )
 
     add_model_arguments(parser)

egs/librispeech/ASR/pruned_stateless_emformer_rnnt2/train.py

@@ -209,7 +209,7 @@ def get_parser():
         "--context-size",
         type=int,
         default=2,
-        help="The context size in the decoder. 1 means bigram; " "2 means tri-gram",
+        help="The context size in the decoder. 1 means bigram; 2 means tri-gram",
     )
 
     parser.add_argument(
@@ -232,7 +232,7 @@ def get_parser():
         "--am-scale",
         type=float,
         default=0.0,
-        help="The scale to smooth the loss with am (output of encoder network)" "part.",
+        help="The scale to smooth the loss with am (output of encoder network) part.",
     )
 
     parser.add_argument(
@@ -898,7 +898,7 @@ def run(rank, world_size, args):
         # the threshold
         if c.duration < 1.0 or c.duration > 20.0:
             logging.warning(
-                f"Exclude cut with ID {c.id} from training. " f"Duration: {c.duration}"
+                f"Exclude cut with ID {c.id} from training. Duration: {c.duration}"
             )
             return False
 

egs/librispeech/ASR/pruned_transducer_stateless/decode.py

@@ -265,7 +265,7 @@ def get_parser():
         "--context-size",
         type=int,
         default=2,
-        help="The context size in the decoder. 1 means bigram; " "2 means tri-gram",
+        help="The context size in the decoder. 1 means bigram; 2 means tri-gram",
     )
     parser.add_argument(
         "--max-sym-per-frame",
@@ -703,7 +703,7 @@ def main():
         ]
         if len(filenames) == 0:
             raise ValueError(
-                f"No checkpoints found for" f" --iter {params.iter}, --avg {params.avg}"
+                f"No checkpoints found for --iter {params.iter}, --avg {params.avg}"
             )
         elif len(filenames) < params.avg:
             raise ValueError(

egs/librispeech/ASR/pruned_transducer_stateless/export.py

@@ -105,7 +105,7 @@ def get_parser():
         "--context-size",
         type=int,
         default=2,
-        help="The context size in the decoder. 1 means bigram; " "2 means tri-gram",
+        help="The context size in the decoder. 1 means bigram; 2 means tri-gram",
     )
 
     parser.add_argument(

egs/librispeech/ASR/pruned_transducer_stateless/pretrained.py

@@ -168,7 +168,7 @@ def get_parser():
         "--context-size",
         type=int,
         default=2,
-        help="The context size in the decoder. 1 means bigram; " "2 means tri-gram",
+        help="The context size in the decoder. 1 means bigram; 2 means tri-gram",
     )
     parser.add_argument(
         "--max-sym-per-frame",
@@ -220,9 +220,9 @@ def read_sound_files(
     ans = []
     for f in filenames:
         wave, sample_rate = torchaudio.load(f)
-        assert sample_rate == expected_sample_rate, (
-            f"expected sample rate: {expected_sample_rate}. " f"Given: {sample_rate}"
-        )
+        assert (
+            sample_rate == expected_sample_rate
+        ), f"expected sample rate: {expected_sample_rate}. Given: {sample_rate}"
         # We use only the first channel
         ans.append(wave[0])
     return ans

egs/librispeech/ASR/pruned_transducer_stateless/streaming_decode.py

@@ -158,7 +158,7 @@ def get_parser():
         "--context-size",
         type=int,
         default=2,
-        help="The context size in the decoder. 1 means bigram; " "2 means tri-gram",
+        help="The context size in the decoder. 1 means bigram; 2 means tri-gram",
     )
 
     parser.add_argument(
@@ -519,7 +519,7 @@ def main():
         ]
         if len(filenames) == 0:
             raise ValueError(
-                f"No checkpoints found for" f" --iter {params.iter}, --avg {params.avg}"
+                f"No checkpoints found for --iter {params.iter}, --avg {params.avg}"
             )
         elif len(filenames) < params.avg:
             raise ValueError(

egs/librispeech/ASR/pruned_transducer_stateless/train.py

@@ -203,7 +203,7 @@ def get_parser():
         "--context-size",
         type=int,
         default=2,
-        help="The context size in the decoder. 1 means bigram; " "2 means tri-gram",
+        help="The context size in the decoder. 1 means bigram; 2 means tri-gram",
     )
 
     parser.add_argument(
@@ -226,7 +226,7 @@ def get_parser():
         "--am-scale",
         type=float,
         default=0.0,
-        help="The scale to smooth the loss with am (output of encoder network)" "part.",
+        help="The scale to smooth the loss with am (output of encoder network) part.",
     )
 
     parser.add_argument(
@@ -889,7 +889,7 @@ def run(rank, world_size, args):
         # the threshold
         if c.duration < 1.0 or c.duration > 20.0:
             logging.warning(
-                f"Exclude cut with ID {c.id} from training. " f"Duration: {c.duration}"
+                f"Exclude cut with ID {c.id} from training. Duration: {c.duration}"
             )
             return False
 

egs/librispeech/ASR/pruned_transducer_stateless2/decode.py

@@ -271,7 +271,7 @@ def get_parser():
         "--context-size",
         type=int,
         default=2,
-        help="The context size in the decoder. 1 means bigram; " "2 means tri-gram",
+        help="The context size in the decoder. 1 means bigram; 2 means tri-gram",
     )
 
     parser.add_argument(
@@ -725,7 +725,7 @@ def main():
         ]
         if len(filenames) == 0:
             raise ValueError(
-                f"No checkpoints found for" f" --iter {params.iter}, --avg {params.avg}"
+                f"No checkpoints found for --iter {params.iter}, --avg {params.avg}"
             )
         elif len(filenames) < params.avg:
             raise ValueError(

egs/librispeech/ASR/pruned_transducer_stateless2/export.py

@@ -116,7 +116,7 @@ def get_parser():
         "--context-size",
         type=int,
         default=2,
-        help="The context size in the decoder. 1 means bigram; " "2 means tri-gram",
+        help="The context size in the decoder. 1 means bigram; 2 means tri-gram",
     )
 
     parser.add_argument(
@@ -168,7 +168,7 @@ def main():
         ]
         if len(filenames) == 0:
             raise ValueError(
-                f"No checkpoints found for" f" --iter {params.iter}, --avg {params.avg}"
+                f"No checkpoints found for --iter {params.iter}, --avg {params.avg}"
             )
         elif len(filenames) < params.avg:
             raise ValueError(

egs/librispeech/ASR/pruned_transducer_stateless2/pretrained.py

@@ -168,7 +168,7 @@ def get_parser():
         "--context-size",
         type=int,
         default=2,
-        help="The context size in the decoder. 1 means bigram; " "2 means tri-gram",
+        help="The context size in the decoder. 1 means bigram; 2 means tri-gram",
     )
     parser.add_argument(
         "--max-sym-per-frame",
@@ -221,9 +221,9 @@ def read_sound_files(
     ans = []
     for f in filenames:
         wave, sample_rate = torchaudio.load(f)
-        assert sample_rate == expected_sample_rate, (
-            f"expected sample rate: {expected_sample_rate}. " f"Given: {sample_rate}"
-        )
+        assert (
+            sample_rate == expected_sample_rate
+        ), f"expected sample rate: {expected_sample_rate}. Given: {sample_rate}"
         # We use only the first channel
         ans.append(wave[0])
     return ans

egs/librispeech/ASR/pruned_transducer_stateless2/streaming_decode.py

@@ -158,7 +158,7 @@ def get_parser():
         "--context-size",
         type=int,
         default=2,
-        help="The context size in the decoder. 1 means bigram; " "2 means tri-gram",
+        help="The context size in the decoder. 1 means bigram; 2 means tri-gram",
     )
 
     parser.add_argument(
@@ -522,7 +522,7 @@ def main():
         ]
         if len(filenames) == 0:
             raise ValueError(
-                f"No checkpoints found for" f" --iter {params.iter}, --avg {params.avg}"
+                f"No checkpoints found for --iter {params.iter}, --avg {params.avg}"
             )
         elif len(filenames) < params.avg:
             raise ValueError(

egs/librispeech/ASR/pruned_transducer_stateless2/train.py

@@ -208,7 +208,7 @@ def get_parser():
         "--initial-lr",
         type=float,
         default=0.003,
-        help="The initial learning rate.  This value should not need to " "be changed.",
+        help="The initial learning rate.  This value should not need to be changed.",
     )
 
     parser.add_argument(
@@ -231,7 +231,7 @@ def get_parser():
         "--context-size",
         type=int,
         default=2,
-        help="The context size in the decoder. 1 means bigram; " "2 means tri-gram",
+        help="The context size in the decoder. 1 means bigram; 2 means tri-gram",
     )
 
     parser.add_argument(
@@ -254,7 +254,7 @@ def get_parser():
         "--am-scale",
         type=float,
         default=0.0,
-        help="The scale to smooth the loss with am (output of encoder network)" "part.",
+        help="The scale to smooth the loss with am (output of encoder network) part.",
     )
 
     parser.add_argument(
@@ -947,7 +947,7 @@ def run(rank, world_size, args):
         # the threshold
         if c.duration < 1.0 or c.duration > 20.0:
             logging.warning(
-                f"Exclude cut with ID {c.id} from training. " f"Duration: {c.duration}"
+                f"Exclude cut with ID {c.id} from training. Duration: {c.duration}"
             )
             return False
 

egs/librispeech/ASR/pruned_transducer_stateless3/decode-giga.py

@@ -188,7 +188,7 @@ def get_parser():
         "--context-size",
         type=int,
         default=2,
-        help="The context size in the decoder. 1 means bigram; " "2 means tri-gram",
+        help="The context size in the decoder. 1 means bigram; 2 means tri-gram",
     )
     parser.add_argument(
         "--max-sym-per-frame",
@@ -552,7 +552,7 @@ def main():
         ]
         if len(filenames) == 0:
             raise ValueError(
-                f"No checkpoints found for" f" --iter {params.iter}, --avg {params.avg}"
+                f"No checkpoints found for --iter {params.iter}, --avg {params.avg}"
             )
         elif len(filenames) < params.avg:
             raise ValueError(

egs/librispeech/ASR/pruned_transducer_stateless3/decode.py

@@ -261,7 +261,7 @@ def get_parser():
         "--context-size",
         type=int,
         default=2,
-        help="The context size in the decoder. 1 means bigram; " "2 means tri-gram",
+        help="The context size in the decoder. 1 means bigram; 2 means tri-gram",
     )
     parser.add_argument(
         "--max-sym-per-frame",
@@ -681,7 +681,7 @@ def decode_one_batch(
         return {key: hyps}
     else:
         return {
-            (f"beam_size_{params.beam_size}_" f"temperature_{params.temperature}"): hyps
+            (f"beam_size_{params.beam_size}_temperature_{params.temperature}"): hyps
         }
 
 
@@ -963,7 +963,7 @@ def main():
         ]
         if len(filenames) == 0:
             raise ValueError(
-                f"No checkpoints found for" f" --iter {params.iter}, --avg {params.avg}"
+                f"No checkpoints found for --iter {params.iter}, --avg {params.avg}"
             )
         elif len(filenames) < params.avg:
             raise ValueError(

egs/librispeech/ASR/pruned_transducer_stateless3/export.py

@@ -231,7 +231,7 @@ def get_parser():
         "--context-size",
         type=int,
         default=2,
-        help="The context size in the decoder. 1 means bigram; " "2 means tri-gram",
+        help="The context size in the decoder. 1 means bigram; 2 means tri-gram",
     )
 
     parser.add_argument(
@@ -607,7 +607,7 @@ def main():
         ]
         if len(filenames) == 0:
             raise ValueError(
-                f"No checkpoints found for" f" --iter {params.iter}, --avg {params.avg}"
+                f"No checkpoints found for --iter {params.iter}, --avg {params.avg}"
             )
         elif len(filenames) < params.avg:
             raise ValueError(

egs/librispeech/ASR/pruned_transducer_stateless3/jit_pretrained.py

@@ -142,9 +142,9 @@ def read_sound_files(
     ans = []
     for f in filenames:
         wave, sample_rate = torchaudio.load(f)
-        assert sample_rate == expected_sample_rate, (
-            f"expected sample rate: {expected_sample_rate}. " f"Given: {sample_rate}"
-        )
+        assert (
+            sample_rate == expected_sample_rate
+        ), f"expected sample rate: {expected_sample_rate}. Given: {sample_rate}"
         # We use only the first channel
         ans.append(wave[0])
     return ans

egs/librispeech/ASR/pruned_transducer_stateless3/onnx_pretrained.py

@@ -140,9 +140,9 @@ def read_sound_files(
     ans = []
     for f in filenames:
         wave, sample_rate = torchaudio.load(f)
-        assert sample_rate == expected_sample_rate, (
-            f"expected sample rate: {expected_sample_rate}. " f"Given: {sample_rate}"
-        )
+        assert (
+            sample_rate == expected_sample_rate
+        ), f"expected sample rate: {expected_sample_rate}. Given: {sample_rate}"
         # We use only the first channel
         ans.append(wave[0])
     return ans