fix comments

egs/tedlium3/ASR/local/compute_fbank_tedlium.py

@@ -70,7 +70,7 @@ def compute_fbank_tedlium():
             cut_set = CutSet.from_manifests(
                 recordings=m["recordings"],
                 supervisions=m["supervisions"],
-            ).trim_to_supervisions(keep_overlapping=False)
+            )
             if "train" in partition:
                 cut_set = (
                     cut_set
@@ -85,6 +85,8 @@ def compute_fbank_tedlium():
                 executor=ex,
                 storage_type=ChunkedLilcomHdf5Writer,
             )
+            # Split long cuts into many short and un-overlapping cuts
+            cut_set = cut_set.trim_to_supervisions(keep_overlapping=False)
             cut_set.to_json(output_dir / f"cuts_{partition}.json.gz")
 
 

egs/tedlium3/ASR/local/convert_transcript_words_to_bpe_ids.py

@@ -42,7 +42,7 @@ def convert_texts_into_ids(
     texts: List[str],
     unk_id: int,
     sp: spm.SentencePieceProcessor,
-) -> List[int]:
+) -> List[List[int]]:
     """
     Args:
       texts:
@@ -50,7 +50,7 @@ def convert_texts_into_ids(
       unk_id:
         A number id for the token '<unk>'.
     Returns:
-      Return a integer list of bpe ids.
+      Return an integer list of bpe ids.
     """
     y = []
     for text in texts:

egs/tedlium3/ASR/local/prepare_transcripts.py

@@ -0,0 +1,95 @@
+#!/usr/bin/env python3
+# Copyright    2021  Xiaomi Corp.        (authors: Mingshuang Luo)
+#
+# See ../../../../LICENSE for clarification regarding multiple authors
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+
+"""
+This script takes as input supervisions json dir "data/manifests"
+consisting of supervisions_TRAIN.json and does the following:
+
+1. Generate train.text.
+
+"""
+import argparse
+import json
+import logging
+from pathlib import Path
+
+
+def get_args():
+    parser = argparse.ArgumentParser()
+    parser.add_argument(
+        "--manifests-dir",
+        type=str,
+        help="""Input directory.
+        """,
+    )
+    parser.add_argument(
+        "--lang-dir",
+        type=str,
+        help="""Output directory.
+        """,
+    )
+
+    return parser.parse_args()
+
+
+def prepare_transcripts(manifests_dir: str, lang_dir: str):
+    """
+    Args:
+      manifests_dir:
+        The manifests directory, e.g., data/manifests.
+      lang_dir:
+        The language directory, e.g., data/lang_phone.
+
+    Return:
+      The train.text in lang_dir.
+    """
+    texts = []
+
+    supervisions_train = Path(manifests_dir) / "supervisions_train.json"
+    train_text = Path(lang_dir) / "train.text"
+
+    logging.info(f"Loading {supervisions_train}!")
+    with open(supervisions_train, "r") as load_f:
+        load_dicts = json.load(load_f)
+        for load_dict in load_dicts:
+            text = load_dict["text"]
+            texts.append(text)
+
+    with open(train_text, "w") as f:
+        for text in texts:
+            f.write(text)
+            f.write("\n")
+
+
+def main():
+    args = get_args()
+    manifests_dir = Path(args.manifests_dir)
+    lang_dir = Path(args.lang_dir)
+
+    logging.info("Generating train.text")
+    prepare_transcripts(manifests_dir, lang_dir)
+
+
+if __name__ == "__main__":
+    formatter = (
+        "%(asctime)s %(levelname)s [%(filename)s:%(lineno)d] %(message)s"
+    )
+
+    logging.basicConfig(format=formatter, level=logging.INFO)
+
+    main()

egs/tedlium3/ASR/prepare.sh

@@ -71,13 +71,14 @@ fi
 if [ $stage -le 0 ] && [ $stop_stage -ge 0 ]; then
   log "Stage 0: Download data"
 
-  # If you have pre-downloaded it to /path/to/LibriSpeech,
+  # If you have pre-downloaded it to /path/to/tedlium3,
   # you can create a symlink
   #
   # ln -sfv /path/to/tedlium3 $dl_dir/tedlium3
   #
-  if [ ! -d $dl_dir/tedlium ]; then
+  if [ ! -d $dl_dir/tedlium3 ]; then
     lhotse download tedlium $dl_dir
+    mv $dl_dir/TEDLIUM_release-3 $dl_dir/tedlium3
   fi
 
   # If you have pre-downloaded it to /path/to/musan,
@@ -127,13 +128,13 @@ if [ $stage -le 5 ] && [ $stop_stage -ge 5 ]; then
     ./local/prepare_transcripts.py \
       --lang-dir $lang_dir \
       --manifests-dir data/manifests
-
-  cat download/tedlium3/TEDLIUM.152k.dic |
-  grep -v -w "<s>" |
-  grep -v -w "</s>" |
-  grep -v -w "<unk>" |
-  LANG= LC_ALL= sort |
-  sed 's:([0-9])::g' > $lang_dir/lexicon_words.txt
+  fi
+  cat download/tedlium3/TEDLIUM.152k.dic | \
+    grep -v -w "<s>" | \
+    grep -v -w "</s>" | \
+    grep -v -w "<unk>" | \
+    LANG= LC_ALL= sort | \
+    sed 's:([0-9])::g' > $lang_dir/lexicon_words.txt
 
   (echo '<UNK> <UNK>'; ) |
     cat - $lang_dir/lexicon_words.txt |
@@ -174,69 +175,3 @@ if [ $stage -le 6 ] && [ $stop_stage -ge 6 ]; then
     fi
   done
 fi
-
-if [ $stage -le 7 ] && [ $stop_stage -ge 7 ]; then
-  log "Stage 7: Prepare bigram P"
-
-  for vocab_size in ${vocab_sizes[@]}; do
-    lang_dir=data/lang_bpe_${vocab_size}
-
-    if [ ! -f $lang_dir/transcript_tokens.txt ]; then
-      ./local/convert_transcript_words_to_tokens.py \
-        --lexicon $lang_dir/lexicon.txt \
-        --transcript $lang_dir/transcript_words.txt \
-        --oov "<UNK>" \
-        > $lang_dir/transcript_tokens.txt
-    fi
-
-    if [ ! -f $lang_dir/P.arpa ]; then
-      ./shared/make_kn_lm.py \
-        -ngram-order 2 \
-        -text $lang_dir/transcript_tokens.txt \
-        -lm $lang_dir/P.arpa
-    fi
-
-    if [ ! -f $lang_dir/P.fst.txt ]; then
-      python3 -m kaldilm \
-        --read-symbol-table="$lang_dir/tokens.txt" \
-        --disambig-symbol='#0' \
-        --max-order=2 \
-        $lang_dir/P.arpa > $lang_dir/P.fst.txt
-    fi
-  done
-fi
-
-if [ $stage -le 8 ] && [ $stop_stage -ge 8 ]; then
-  log "Stage 8: Prepare G"
-  # We assume you have install kaldilm, if not, please install
-  # it using: pip install kaldilm
-
-  mkdir -p data/lm
-  if [ ! -f data/lm/G_3_gram.fst.txt ]; then
-    # It is used in building HLG
-    python3 -m kaldilm \
-      --read-symbol-table="data/lang_phone/words.txt" \
-      --disambig-symbol='#0' \
-      --max-order=3 \
-      data/lm/lm_3_gram.arpa > data/lm/G_3_gram.fst.txt
-  fi
-
-  if [ ! -f data/lm/G_4_gram.fst.txt ]; then
-    # It is used for LM rescoring
-    python3 -m kaldilm \
-      --read-symbol-table="data/lang_phone/words.txt" \
-      --disambig-symbol='#0' \
-      --max-order=4 \
-      data/lm/lm_4_gram.arpa > data/lm/G_4_gram.fst.txt
-  fi
-fi
-echo 'completing the G building....'
-if [ $stage -le 9 ] && [ $stop_stage -ge 9 ]; then
-  log "Stage 9: Compile HLG"
-  ./local/compile_hlg.py --lang-dir data/lang_phone
-
-  for vocab_size in ${vocab_sizes[@]}; do
-    lang_dir=data/lang_bpe_${vocab_size}
-    ./local/compile_hlg.py --lang-dir $lang_dir
-  done
-fi

egs/tedlium3/ASR/transducer_stateless/README.md

@@ -16,5 +16,5 @@ export CUDA_VISIBLE_DEVICES="0,1,2,3"
   --num-epochs 30 \
   --start-epoch 0 \
   --exp-dir transducer_stateless/exp \
-  --max-duration 200 \
+  --max-duration 200
 ```

egs/tedlium3/ASR/transducer_stateless/decode.py

@@ -34,6 +34,15 @@ Usage:
         --max-duration 100 \
         --decoding-method beam_search \
         --beam-size 4
+
+(3) modified beam search
+./transducer_stateless/decode.py \
+        --epoch 29 \
+        --avg 16 \
+        --exp-dir ./transducer_stateless/exp \
+        --max-duration 100 \
+        --decoding-method modified_beam_search \
+        --beam-size 4
 """
 
 

egs/tedlium3/ASR/transducer_stateless/export.py

@@ -39,7 +39,7 @@ To use the generated file with `transducer_stateless/decode.py`, you can do:
         --exp-dir ./transducer_stateless/exp \
         --epoch 9999 \
         --avg 1 \
-        --max-duration 1 \
+        --max-duration 100 \
         --bpe-model data/lang_bpe_500/bpe.model
 """
 

egs/tedlium3/ASR/transducer_stateless/pretrained.py

@@ -25,7 +25,7 @@ Usage:
         --method greedy_search \
         --max-sym-per-frame 1 \
         /path/to/foo.wav \
-        /path/to/bar.wav \
+        /path/to/bar.wav
 
 (2) beam search
 ./transducer_stateless/pretrained.py \
@@ -34,7 +34,7 @@ Usage:
         --method beam_search \
         --beam-size 4 \
         /path/to/foo.wav \
-        /path/to/bar.wav \
+        /path/to/bar.wav
 
 (3) modified beam search
 ./transducer_stateless/pretrained.py \
@@ -43,7 +43,7 @@ Usage:
         --method modified_beam_search \
         --beam-size 4 \
         /path/to/foo.wav \
-        /path/to/bar.wav \
+        /path/to/bar.wav
 
 You can also use `./transducer_stateless/exp/epoch-xx.pt`.
 

egs/tedlium3/ASR/transducer_stateless/train.py

@@ -397,7 +397,6 @@ def compute_loss(
     feature_lens = supervisions["num_frames"].to(device)
 
     texts = batch["supervisions"]["text"]
-
     unk_id = params.unk_id
     y = convert_texts_into_ids(texts, unk_id, sp=sp)
     y = k2.RaggedTensor(y).to(device)