diff --git a/egs/bengaliai_speech/ASR/local/preprocess_bengaliai_speech.py b/egs/bengaliai_speech/ASR/local/preprocess_bengaliai_speech.py
new file mode 100755
index 000000000..fbcfa3ce5
--- /dev/null
+++ b/egs/bengaliai_speech/ASR/local/preprocess_bengaliai_speech.py
@@ -0,0 +1,124 @@
+#!/usr/bin/env python3
+# Copyright    2023  Xiaomi Corp.        (authors: Yifan Yang)
+#
+# 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.
+
+import argparse
+import logging
+import re
+from pathlib import Path
+from typing import Optional
+
+from lhotse import CutSet, SupervisionSegment
+from lhotse.recipes.utils import read_manifests_if_cached
+
+
+def get_args():
+    parser = argparse.ArgumentParser()
+
+    parser.add_argument(
+        "--dataset",
+        type=str,
+        help="""Dataset parts to compute fbank. If None, we will use all""",
+    )
+
+    return parser.parse_args()
+
+
+def normalize_text(utt: str) -> str:
+    punc = '~`!#$%^&*()_+-=|\';":/.,?><~·！@#￥%……&*（）——+-=“：’；、。，？》《{}'
+    return re.sub(r"[{0}]+".format(punc), "", utt).upper()
+
+
+def preprocess_bengaliai_speech(
+    dataset: Optional[str] = None,
+):
+    src_dir = Path(f"data/manifests")
+    output_dir = Path(f"data/fbank")
+    output_dir.mkdir(exist_ok=True)
+
+    if dataset is None:
+        dataset_parts = (
+            "train",
+            "valid",
+            "test",
+        )
+    else:
+        dataset_parts = dataset.split(" ", -1)
+
+    logging.info("Loading manifest")
+    prefix = f"bengaliai_speech"
+    suffix = "jsonl.gz"
+    manifests = read_manifests_if_cached(
+        dataset_parts=dataset_parts,
+        output_dir=src_dir,
+        suffix=suffix,
+        prefix=prefix,
+    )
+    assert manifests is not None
+
+    assert len(manifests) == len(dataset_parts), (
+        len(manifests),
+        len(dataset_parts),
+        list(manifests.keys()),
+        dataset_parts,
+    )
+
+    for partition, m in manifests.items():
+        logging.info(f"Processing {partition}")
+        raw_cuts_path = output_dir / f"{prefix}_cuts_{partition}_raw.{suffix}"
+        if raw_cuts_path.is_file():
+            logging.info(f"{partition} already exists - skipping")
+            continue
+
+        logging.info(f"Normalizing text in {partition}")
+        for sup in m["supervisions"]:
+            if sup.text is None:
+                continue
+            text = str(sup.text)
+            orig_text = text
+            sup.text = normalize_text(sup.text)
+            text = str(sup.text)
+            if len(orig_text) != len(text):
+                logging.info(
+                    f"\nOriginal text vs normalized text:\n{orig_text}\n{text}"
+                )
+
+        # Create long-recording cut manifests.
+        cut_set = CutSet.from_manifests(
+            recordings=m["recordings"],
+            supervisions=m["supervisions"],
+        ).resample(16000)
+
+        # Run data augmentation that needs to be done in the
+        # time domain.
+        logging.info(f"Saving to {raw_cuts_path}")
+        cut_set.to_file(raw_cuts_path)
+
+
+def main():
+    formatter = "%(asctime)s %(levelname)s [%(filename)s:%(lineno)d] %(message)s"
+
+    logging.basicConfig(format=formatter, level=logging.INFO)
+    args = get_args()
+    logging.info(vars(args))
+    preprocess_bengaliai_speech(
+        dataset=args.dataset,
+    )
+    logging.info("Done")
+
+
+if __name__ == "__main__":
+    main()
diff --git a/egs/bengaliai_speech/ASR/prepare.sh b/egs/bengaliai_speech/ASR/prepare.sh
new file mode 100755
index 000000000..45931abc4
--- /dev/null
+++ b/egs/bengaliai_speech/ASR/prepare.sh
@@ -0,0 +1,233 @@
+#!/usr/bin/env bash
+
+set -eou pipefail
+
+nj=32
+stage=-1
+stop_stage=100
+
+# Split data/set to a number of pieces
+# This is to avoid OOM during feature extraction.
+num_per_split=4000
+
+# We assume dl_dir (download dir) contains the following
+# directories and files. If not, they will be downloaded
+# by this script automatically.
+#
+#  - $dl_dir/bengaliai_speech
+#      This directory contains the following files downloaded by
+#       kaggle competitions download -c bengaliai-speech
+# 
+#     - train_mp3s
+#     - test_mp3s
+#     - examples
+#     - train.csv
+#     - sample_submission.csv
+#
+#  - $dl_dir/musan
+#      This directory contains the following directories downloaded from
+#       http://www.openslr.org/17/
+#
+#     - music
+#     - noise
+#     - speech
+
+dl_dir=$PWD/download
+
+. shared/parse_options.sh || exit 1
+
+# vocab size for sentence piece models.
+# It will generate data/lang_bpe_xxx,
+# data/lang_bpe_yyy if the array contains xxx, yyy
+vocab_sizes=(
+  # 5000
+  # 2000
+  # 1000
+  500
+)
+
+# All files generated by this script are saved in "data".
+# You can safely remove "data" and rerun this script to regenerate it.
+mkdir -p data
+
+log() {
+  # This function is from espnet
+  local fname=${BASH_SOURCE[1]##*/}
+  echo -e "$(date '+%Y-%m-%d %H:%M:%S') (${fname}:${BASH_LINENO[0]}:${FUNCNAME[1]}) $*"
+}
+
+log "dl_dir: $dl_dir"
+
+if [ $stage -le 0 ] && [ $stop_stage -ge 0 ]; then
+  log "Stage 0: Download data"
+
+  # If you have pre-downloaded it to /path/to/bengaliai_speech,
+  # you can create a symlink
+  #
+  #   ln -sfv /path/to/bengaliai_speech $dl_dir/bengaliai_speech
+  #
+  if [ ! -d $dl_dir/bengaliai/train_mp3s ]; then
+    kaggle competitions download -c bengaliai-speech
+  fi
+
+  # If you have pre-downloaded it to /path/to/musan,
+  # you can create a symlink
+  #
+  #   ln -sfv /path/to/musan $dl_dir/
+  #
+  if [ ! -d $dl_dir/musan ]; then
+    lhotse download musan $dl_dir
+  fi
+fi
+
+if [ $stage -le 1 ] && [ $stop_stage -ge 1 ]; then
+  log "Stage 1: Prepare Bengali.AI Speech manifest"
+  # We assume that you have downloaded the Bengali.AI Speech corpus
+  # to $dl_dir/bengaliai_speech
+  mkdir -p data/manifests
+  if [ ! -e data/manifests/.bengaliai_speech.done ]; then
+    lhotse prepare bengaliai-speech -j $nj $dl_dir/bengaliai_speech data/manifests
+    touch data/manifests/.bengaliai_speech.done
+  fi
+fi
+
+if [ $stage -le 2 ] && [ $stop_stage -ge 2 ]; then
+  log "Stage 2: Prepare musan manifest"
+  # We assume that you have downloaded the musan corpus
+  # to data/musan
+  mkdir -p data/manifests
+  if [ ! -e data/manifests/.musan.done ]; then
+    lhotse prepare musan $dl_dir/musan data/manifests
+    touch data/manifests/.musan.done
+  fi
+fi
+
+if [ $stage -le 3 ] && [ $stop_stage -ge 3 ]; then
+  log "Stage 3: Preprocess Bengali.AI Speech manifest"
+  mkdir -p data/fbank
+  if [ ! -e data/fbank/.preprocess_complete ]; then
+    ./local/preprocess_bengaliai_speech.py
+    touch data/fbank/.preprocess_complete
+  fi
+fi
+
+if [ $stage -le 4 ] && [ $stop_stage -ge 4 ]; then
+  log "Stage 4: Compute fbank for valid and test subsets of Bengali.AI Speech"
+  if [ ! -e data/fbank/.bengaliai_speech_valid_test.done ]; then
+    ./local/compute_fbank_bengaliai_speech_valid_test.py
+    touch data/fbank/.bengaliai_speech_valid_test.done
+  fi
+fi
+
+if [ $stage -le 5 ] && [ $stop_stage -ge 5 ]; then
+  log "Stage 5: Split train subset into pieces"
+  split_dir=data/fbank/bengaliai_speech_train_split
+  if [ ! -e $split_dir/.bengaliai_speech_train_split.done ]; then
+    lhotse split-lazy ./data/fbank/bengaliai_speech_cuts_train_raw.jsonl.gz $split_dir $num_per_split
+    touch $split_dir/.bengaliai_speech_train_split.done
+  fi
+fi
+
+if [ $stage -le 6 ] && [ $stop_stage -ge 6 ]; then
+  log "Stage 6: Compute features for train subset of Bengali.AI Speech"
+  if [ ! -e data/fbank/.bengaliai_speech_train.done ]; then
+    ./local/compute_fbank_bengaliai_speech_splits.py \
+      --num-workers $nj \
+      --batch-duration 600 \
+      --start 0 \
+      --num-splits 2000
+    touch data/fbank/.bengaliai_speech_train.done
+  fi
+fi
+
+if [ $stage -le 7 ] && [ $stop_stage -ge 7 ]; then
+  log "Stage 7: Compute fbank for musan"
+  mkdir -p data/fbank
+  if [ ! -e data/fbank/.musan.done ]; then
+    ./local/compute_fbank_musan.py
+    touch data/fbank/.musan.done
+  fi
+fi
+
+if [ $stage -le 8 ] && [ $stop_stage -ge 8 ]; then
+  log "Stage 8: Prepare BPE based lang"
+
+  for vocab_size in ${vocab_sizes[@]}; do
+    lang_dir=data/lang_bpe_${vocab_size}
+    mkdir -p $lang_dir
+
+    if [ ! -f $lang_dir/transcript_words.txt ]; then
+      log "Generate data for BPE training"
+      file=$(
+        find "data/fbank/bengaliai_speech_cuts_dirty_raw.jsonl.gz"
+	find "data/fbank/bengaliai_speech_cuts_dirty_sa_raw.jsonl.gz"
+	find "data/fbank/bengaliai_speech_cuts_clean_raw.jsonl.gz"
+	find "data/fbank/bengaliai_speech_cuts_clean_sa_raw.jsonl.gz"
+      )
+      gunzip -c ${file} | awk -F '"' '{print $30}' > $lang_dir/transcript_words.txt
+
+      # Ensure space only appears once
+      sed -i 's/\t/ /g' $lang_dir/transcript_words.txt
+      sed -i 's/ +/ /g' $lang_dir/transcript_words.txt
+    fi
+ 
+    if [ ! -f $lang_dir/words.txt ]; then
+      cat $lang_dir/transcript_words.txt | sed 's/ /\n/g' \
+        | sort -u | sed '/^$/d' > $lang_dir/words.txt
+      (echo '!SIL'; echo '<SPOKEN_NOISE>'; echo '<UNK>'; ) |
+        cat - $lang_dir/words.txt | sort | uniq | awk '
+        BEGIN {
+          print "<eps> 0";
+        }
+        {
+          if ($1 == "<s>") {
+            print "<s> is in the vocabulary!" | "cat 1>&2"
+            exit 1;
+          }
+          if ($1 == "</s>") {
+            print "</s> is in the vocabulary!" | "cat 1>&2"
+            exit 1;
+          }
+          printf("%s %d\n", $1, NR);
+        }
+        END {
+          printf("#0 %d\n", NR+1);
+          printf("<s> %d\n", NR+2);
+          printf("</s> %d\n", NR+3);
+        }' > $lang_dir/words || exit 1;
+      mv $lang_dir/words $lang_dir/words.txt
+    fi
+ 
+    if [ ! -f $lang_dir/bpe.model ]; then
+      ./local/train_bpe_model.py \
+        --lang-dir $lang_dir \
+        --vocab-size $vocab_size \
+        --transcript $lang_dir/transcript_words.txt
+    fi
+  
+    if [ ! -f $lang_dir/L_disambig.pt ]; then
+      ./local/prepare_lang_bpe.py --lang-dir $lang_dir
+
+      log "Validating $lang_dir/lexicon.txt"
+      ./local/validate_bpe_lexicon.py \
+        --lexicon $lang_dir/lexicon.txt \
+        --bpe-model $lang_dir/bpe.model
+    fi
+
+    if [ ! -f $lang_dir/L.fst ]; then
+      log "Converting L.pt to L.fst"
+      ./shared/convert-k2-to-openfst.py \
+        --olabels aux_labels \
+        $lang_dir/L.pt \
+        $lang_dir/L.fst
+    fi
+
+    if [ ! -f $lang_dir/L_disambig.fst ]; then
+      log "Converting L_disambig.pt to L_disambig.fst"
+      ./shared/convert-k2-to-openfst.py \
+        --olabels aux_labels \
+        $lang_dir/L_disambig.pt \
+        $lang_dir/L_disambig.fst
+    fi
+  done
+fi
diff --git a/egs/bengaliai_speech/ASR/shared b/egs/bengaliai_speech/ASR/shared
new file mode 120000
index 000000000..4cbd91a7e
--- /dev/null
+++ b/egs/bengaliai_speech/ASR/shared
@@ -0,0 +1 @@
+../../../icefall/shared
\ No newline at end of file