[Ready] [Recipes] add aishell2

egs/aishell2/ASR/local/compute_fbank_aishell2.py

@@ -0,0 +1,114 @@
+#!/usr/bin/env python3
+# Copyright    2021  Xiaomi Corp.        (authors: Fangjun Kuang)
+#
+# 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 file computes fbank features of the aishell dataset.
+It looks for manifests in the directory data/manifests.
+
+The generated fbank features are saved in data/fbank.
+"""
+
+import argparse
+import logging
+import os
+from pathlib import Path
+
+import torch
+from lhotse import CutSet, Fbank, FbankConfig, LilcomChunkyWriter
+from lhotse.recipes.utils import read_manifests_if_cached
+
+from icefall.utils import get_executor
+
+# Torch's multithreaded behavior needs to be disabled or
+# it wastes a lot of CPU and slow things down.
+# Do this outside of main() in case it needs to take effect
+# even when we are not invoking the main (e.g. when spawning subprocesses).
+torch.set_num_threads(1)
+torch.set_num_interop_threads(1)
+
+
+def compute_fbank_aishell2(num_mel_bins: int = 80):
+    src_dir = Path("data/manifests")
+    output_dir = Path("data/fbank")
+    num_jobs = min(15, os.cpu_count())
+
+    dataset_parts = (
+        "train",
+        "dev",
+        "test",
+    )
+    prefix = "aishell2"
+    suffix = "jsonl.gz"
+    manifests = read_manifests_if_cached(
+        dataset_parts=dataset_parts,
+        output_dir=src_dir,
+        prefix=prefix,
+        suffix=suffix,
+    )
+    assert manifests is not None
+
+    extractor = Fbank(FbankConfig(num_mel_bins=num_mel_bins))
+
+    with get_executor() as ex:  # Initialize the executor only once.
+        for partition, m in manifests.items():
+            if (output_dir / f"{prefix}_cuts_{partition}.{suffix}").is_file():
+                logging.info(f"{partition} already exists - skipping.")
+                continue
+            logging.info(f"Processing {partition}")
+            cut_set = CutSet.from_manifests(
+                recordings=m["recordings"],
+                supervisions=m["supervisions"],
+            )
+            if "train" in partition:
+                cut_set = (
+                    cut_set
+                    + cut_set.perturb_speed(0.9)
+                    + cut_set.perturb_speed(1.1)
+                )
+            cut_set = cut_set.compute_and_store_features(
+                extractor=extractor,
+                storage_path=f"{output_dir}/{prefix}_feats_{partition}",
+                # when an executor is specified, make more partitions
+                num_jobs=num_jobs if ex is None else 80,
+                executor=ex,
+                storage_type=LilcomChunkyWriter,
+            )
+            cut_set.to_file(output_dir / f"{prefix}_cuts_{partition}.{suffix}")
+
+
+def get_args():
+    parser = argparse.ArgumentParser()
+    parser.add_argument(
+        "--num-mel-bins",
+        type=int,
+        default=80,
+        help="""The number of mel bins for Fbank""",
+    )
+
+    return parser.parse_args()
+
+
+if __name__ == "__main__":
+    formatter = (
+        "%(asctime)s %(levelname)s [%(filename)s:%(lineno)d] %(message)s"
+    )
+
+    logging.basicConfig(format=formatter, level=logging.INFO)
+
+    args = get_args()
+    compute_fbank_aishell2(num_mel_bins=args.num_mel_bins)

egs/aishell2/ASR/local/compute_fbank_musan.py

@@ -0,0 +1 @@
+../../../librispeech/ASR/local/compute_fbank_musan.py

egs/aishell2/ASR/local/display_manifest_statistics.py

@@ -0,0 +1,193 @@
+#!/usr/bin/env python3
+# Copyright    2021  Xiaomi Corp.        (authors: Fangjun Kuang)
+#
+# 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 file displays duration statistics of utterances in a manifest.
+You can use the displayed value to choose minimum/maximum duration
+to remove short and long utterances during the training.
+
+See the function `remove_short_and_long_utt()` in transducer_stateless/train.py
+for usage.
+"""
+
+
+from lhotse import load_manifest_lazy
+
+
+def main():
+    #  path = "./data/fbank/aishell2_cuts_train.jsonl.gz"
+    #  path = "./data/fbank/aishell2_cuts_test.jsonl.gz"
+    path = "./data/fbank/aishell2_cuts_dev.jsonl.gz"
+
+    cuts = load_manifest_lazy(path)
+    cuts.describe()
+
+
+if __name__ == "__main__":
+    main()
+
+"""
+## train (after speed perturb)
+Cuts count: 360294
+Total duration (hours): 455.6
+Speech duration (hours): 455.6 (100.0%)
+***
+Duration statistics (seconds):
+mean    4.6
+std     1.4
+min     1.1
+0.1%    1.8
+0.5%    2.2
+1%      2.3
+5%      2.7
+10%     3.0
+10%     3.0
+25%     3.5
+50%     4.3
+75%     5.4
+90%     6.5
+95%     7.2
+99%     8.8
+99.5%   9.4
+99.9%   10.9
+max     16.1
+
+## test
+Cuts count: 7176
+Total duration (hours): 10.0
+Speech duration (hours): 10.0 (100.0%)
+***
+Duration statistics (seconds):
+mean    5.0
+std     1.6
+min     1.9
+0.1%    2.2
+0.5%    2.4
+1%      2.6
+5%      3.0
+10%     3.2
+10%     3.2
+25%     3.8
+50%     4.7
+75%     5.9
+90%     7.3
+95%     8.2
+99%     9.9
+99.5%   10.7
+99.9%   11.9
+max     14.7
+
+## dev
+Cuts count: 14326
+Total duration (hours): 18.1
+Speech duration (hours): 18.1 (100.0%)
+***
+Duration statistics (seconds):
+mean    4.5
+std     1.3
+min     1.6
+0.1%    2.1
+0.5%    2.3
+1%      2.4
+5%      2.9
+10%     3.1
+10%     3.1
+25%     3.5
+50%     4.3
+75%     5.4
+90%     6.4
+95%     7.0
+99%     8.4
+99.5%   8.9
+99.9%   10.3
+max     12.5
+
+## aidatatang_200zh (train)
+Cuts count: 164905
+Total duration (hours): 139.9
+Speech duration (hours): 139.9 (100.0%)
+***
+Duration statistics (seconds):
+mean    3.1
+std     1.1
+min     1.1
+0.1%    1.5
+0.5%    1.7
+1%      1.8
+5%      2.0
+10%     2.1
+10%     2.1
+25%     2.3
+50%     2.7
+75%     3.4
+90%     4.6
+95%     5.4
+99%     7.1
+99.5%   7.8
+99.9%   9.1
+max     16.3
+
+## aidatatang_200zh (test)
+Cuts count: 48144
+Total duration (hours): 40.2
+Speech duration (hours): 40.2 (100.0%)
+***
+Duration statistics (seconds):
+mean    3.0
+std     1.1
+min     0.9
+0.1%    1.5
+0.5%    1.8
+1%      1.8
+5%      2.0
+10%     2.1
+10%     2.1
+25%     2.3
+50%     2.6
+75%     3.4
+90%     4.4
+95%     5.2
+99%     6.9
+99.5%   7.5
+99.9%   9.0
+max     21.8
+
+## aidatatang_200zh (dev)
+Cuts count: 24216
+Total duration (hours): 20.2
+Speech duration (hours): 20.2 (100.0%)
+***
+Duration statistics (seconds):
+mean    3.0
+std     1.0
+min     1.2
+0.1%    1.6
+0.5%    1.7
+1%      1.8
+5%      2.0
+10%     2.1
+10%     2.1
+25%     2.3
+50%     2.7
+75%     3.4
+90%     4.4
+95%     5.1
+99%     6.7
+99.5%   7.3
+99.9%   8.8
+max     11.3
+"""

egs/aishell2/ASR/local/prepare_char.py

@@ -0,0 +1 @@
+../../../aidatatang_200zh/ASR/local/prepare_char.py

egs/aishell2/ASR/local/text2token.py

@@ -0,0 +1 @@
+../../../aidatatang_200zh/ASR/local/text2token.py