From d2fe50416367bd107356a10162a753af6a490786 Mon Sep 17 00:00:00 2001
From: Mingshuang Luo <37799481+luomingshuang@users.noreply.github.com>
Date: Wed, 10 Nov 2021 11:19:20 +0800
Subject: [PATCH] Do some changes

---
 egs/timit/ASR/local/compute_fbank_timit.py | 2 +-
 egs/timit/ASR/local/prepare_lexicon.py     | 2 +-
 egs/timit/ASR/prepare.sh                   | 8 ++++----
 egs/timit/ASR/tdnn_ligru_ctc/decode.py     | 1 -
 egs/timit/ASR/tdnn_ligru_ctc/model.py      | 5 ++---
 egs/timit/ASR/tdnn_lstm_ctc/decode.py      | 1 -
 6 files changed, 8 insertions(+), 11 deletions(-)

diff --git a/egs/timit/ASR/local/compute_fbank_timit.py b/egs/timit/ASR/local/compute_fbank_timit.py
index 4013c36a8..70be7051c 100644
--- a/egs/timit/ASR/local/compute_fbank_timit.py
+++ b/egs/timit/ASR/local/compute_fbank_timit.py
@@ -70,7 +70,7 @@ def compute_fbank_timit():
                 recordings=m["recordings"],
                 supervisions=m["supervisions"],
             )
-            if "TRAIN" in partition:
+            if partition == "TRAIN":
                 cut_set = (
                     cut_set
                     + cut_set.perturb_speed(0.9)
diff --git a/egs/timit/ASR/local/prepare_lexicon.py b/egs/timit/ASR/local/prepare_lexicon.py
index e2d19c643..f0168ebd6 100644
--- a/egs/timit/ASR/local/prepare_lexicon.py
+++ b/egs/timit/ASR/local/prepare_lexicon.py
@@ -58,7 +58,7 @@ def prepare_lexicon(manifests_dir: str, lang_dir: str):
     Return:
       The lexicon.txt file and the train.text in lang_dir.
     """
-    phones = set([])
+    phones = set()
 
     supervisions_train = Path(manifests_dir) / "supervisions_TRAIN.json"
     lexicon = Path(lang_dir) / "lexicon.txt"
diff --git a/egs/timit/ASR/prepare.sh b/egs/timit/ASR/prepare.sh
index e7805b086..ae1b96a68 100644
--- a/egs/timit/ASR/prepare.sh
+++ b/egs/timit/ASR/prepare.sh
@@ -23,8 +23,8 @@ stop_stage=100
 #	     on 39 phones. About how to get these LM files, you can know it 
 #      from https://github.com/luomingshuang/Train_LM_with_kaldilm.
 #	
-#	    - lm_3_gram_tgmed.arpa
-#     - lm_4_gram_tgmed.arpa
+#	    - lm_3_gram.arpa
+#     - lm_4_gram.arpa
 #
 #  - $dl_dir/musan
 #      This directory contains the following directories downloaded from
@@ -135,7 +135,7 @@ if [ $stage -le 6 ] && [ $stop_stage -ge 6 ]; then
       --read-symbol-table="data/lang_phone/words.txt" \
       --disambig-symbol='#0' \
       --max-order=3 \
-      $dl_dir/lm/lm_3_gram_tgmed.arpa > data/lm/G_3_gram.fst.txt
+      $dl_dir/lm/lm_3_gram.arpa > data/lm/G_3_gram.fst.txt
   fi
 
   if [ ! -f data/lm/G_4_gram.fst.txt ]; then
@@ -144,7 +144,7 @@ if [ $stage -le 6 ] && [ $stop_stage -ge 6 ]; then
       --read-symbol-table="data/lang_phone/words.txt" \
       --disambig-symbol='#0' \
       --max-order=4 \
-      $dl_dir/lm/lm_4_gram_tgmed.arpa > data/lm/G_4_gram.fst.txt
+      $dl_dir/lm/lm_4_gram.arpa > data/lm/G_4_gram.fst.txt
   fi
 fi
 
diff --git a/egs/timit/ASR/tdnn_ligru_ctc/decode.py b/egs/timit/ASR/tdnn_ligru_ctc/decode.py
index d5085d32f..b141e58fa 100644
--- a/egs/timit/ASR/tdnn_ligru_ctc/decode.py
+++ b/egs/timit/ASR/tdnn_ligru_ctc/decode.py
@@ -410,7 +410,6 @@ def main():
     if params.method in ["nbest-rescoring", "whole-lattice-rescoring"]:
         if not (params.lm_dir / "G_4_gram.pt").is_file():
             logging.info("Loading G_4_gram.fst.txt")
-            logging.warning("It may take 8 minutes.")
             with open(params.lm_dir / "G_4_gram.fst.txt") as f:
                 first_word_disambig_id = lexicon.word_table["#0"]
 
diff --git a/egs/timit/ASR/tdnn_ligru_ctc/model.py b/egs/timit/ASR/tdnn_ligru_ctc/model.py
index 0a4e6d82b..890f48247 100644
--- a/egs/timit/ASR/tdnn_ligru_ctc/model.py
+++ b/egs/timit/ASR/tdnn_ligru_ctc/model.py
@@ -452,9 +452,8 @@ class LiGRU_Layer(torch.nn.Module):
                 ).data
 
             # Sampling the mask
-            drop_mask = self.drop_masks[
-                self.drop_mask_cnt:self.drop_mask_cnt + self.batch_size
-            ]
+            right_boundary = self.drop_mask_cnt + self.batch_size
+            drop_mask = self.drop_masks[self.drop_mask_cnt:right_boundary]
             self.drop_mask_cnt = self.drop_mask_cnt + self.batch_size
 
         else:
diff --git a/egs/timit/ASR/tdnn_lstm_ctc/decode.py b/egs/timit/ASR/tdnn_lstm_ctc/decode.py
index 41e683779..e9ca96615 100644
--- a/egs/timit/ASR/tdnn_lstm_ctc/decode.py
+++ b/egs/timit/ASR/tdnn_lstm_ctc/decode.py
@@ -409,7 +409,6 @@ def main():
     if params.method in ["nbest-rescoring", "whole-lattice-rescoring"]:
         if not (params.lm_dir / "G_4_gram.pt").is_file():
             logging.info("Loading G_4_gram.fst.txt")
-            logging.warning("It may take 8 minutes.")
             with open(params.lm_dir / "G_4_gram.fst.txt") as f:
                 first_word_disambig_id = lexicon.word_table["#0"]