mirror of
https://github.com/k2-fsa/icefall.git
synced 2025-08-26 10:16:14 +00:00
update datasets for decoding
This commit is contained in:
root
parent
9a5fc2ab71
commit
1d6e80195f
@ -1,247 +0,0 @@
|
||||
# Copyright 2023 Xiaomi Corp. (authors: Zengrui Jin)
|
||||
#
|
||||
# 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 glob
|
||||
import logging
|
||||
import re
|
||||
from pathlib import Path
|
||||
from typing import Dict, List
|
||||
|
||||
import lhotse
|
||||
from lhotse import CutSet, load_manifest_lazy
|
||||
|
||||
|
||||
class MultiDataset:
|
||||
def __init__(self, fbank_dir: str):
|
||||
"""
|
||||
Args:
|
||||
manifest_dir:
|
||||
It is expected to contain the following files:
|
||||
- aishell_cuts_train.jsonl.gz
|
||||
- aishell2_cuts_train.jsonl.gz
|
||||
- aishell4_cuts_train_L.jsonl.gz
|
||||
- aishell4_cuts_train_M.jsonl.gz
|
||||
- aishell4_cuts_train_S.jsonl.gz
|
||||
- alimeeting-far_cuts_train.jsonl.gz
|
||||
- magicdata_cuts_train.jsonl.gz
|
||||
- primewords_cuts_train.jsonl.gz
|
||||
- stcmds_cuts_train.jsonl.gz
|
||||
- thchs_30_cuts_train.jsonl.gz
|
||||
- kespeech/kespeech-asr_cuts_train_phase1.jsonl.gz
|
||||
- kespeech/kespeech-asr_cuts_train_phase2.jsonl.gz
|
||||
- wenetspeech/cuts_L_fixed.jsonl.gz
|
||||
"""
|
||||
self.fbank_dir = Path(fbank_dir)
|
||||
|
||||
def train_cuts(self) -> CutSet:
|
||||
logging.info("About to get multidataset train cuts")
|
||||
|
||||
# THCHS-30
|
||||
logging.info("Loading THCHS-30 in lazy mode")
|
||||
thchs_30_cuts = load_manifest_lazy(
|
||||
self.fbank_dir / "thchs_30_cuts_train.jsonl.gz"
|
||||
)
|
||||
|
||||
# AISHELL-1
|
||||
logging.info("Loading Aishell-1 in lazy mode")
|
||||
aishell_cuts = load_manifest_lazy(
|
||||
self.fbank_dir / "aishell_cuts_train.jsonl.gz"
|
||||
)
|
||||
|
||||
# AISHELL-2
|
||||
logging.info("Loading Aishell-2 in lazy mode")
|
||||
aishell_2_cuts = load_manifest_lazy(
|
||||
self.fbank_dir / "aishell2_cuts_train.jsonl.gz"
|
||||
)
|
||||
|
||||
# AISHELL-4
|
||||
logging.info("Loading Aishell-4 in lazy mode")
|
||||
aishell_4_L_cuts = load_manifest_lazy(
|
||||
self.fbank_dir / "aishell4_cuts_train_L.jsonl.gz"
|
||||
)
|
||||
aishell_4_M_cuts = load_manifest_lazy(
|
||||
self.fbank_dir / "aishell4_cuts_train_M.jsonl.gz"
|
||||
)
|
||||
aishell_4_S_cuts = load_manifest_lazy(
|
||||
self.fbank_dir / "aishell4_cuts_train_S.jsonl.gz"
|
||||
)
|
||||
|
||||
# ST-CMDS
|
||||
logging.info("Loading ST-CMDS in lazy mode")
|
||||
stcmds_cuts = load_manifest_lazy(self.fbank_dir / "stcmds_cuts_train.jsonl.gz")
|
||||
|
||||
# Primewords
|
||||
logging.info("Loading Primewords in lazy mode")
|
||||
primewords_cuts = load_manifest_lazy(
|
||||
self.fbank_dir / "primewords_cuts_train.jsonl.gz"
|
||||
)
|
||||
|
||||
# MagicData
|
||||
logging.info("Loading MagicData in lazy mode")
|
||||
magicdata_cuts = load_manifest_lazy(
|
||||
self.fbank_dir / "magicdata_cuts_train.jsonl.gz"
|
||||
)
|
||||
|
||||
# Ali-Meeting
|
||||
logging.info("Loading Ali-Meeting in lazy mode")
|
||||
alimeeting_cuts = load_manifest_lazy(
|
||||
self.fbank_dir / "alimeeting-far_cuts_train.jsonl.gz"
|
||||
)
|
||||
|
||||
# WeNetSpeech
|
||||
logging.info("Loading WeNetSpeech in lazy mode")
|
||||
wenetspeech_L_cuts = load_manifest_lazy(
|
||||
self.fbank_dir / "wenetspeech" / "cuts_L_fixed.jsonl.gz"
|
||||
)
|
||||
|
||||
# KeSpeech
|
||||
logging.info("Loading KeSpeech in lazy mode")
|
||||
kespeech_1_cuts = load_manifest_lazy(
|
||||
self.fbank_dir / "kespeech" / "kespeech-asr_cuts_train_phase1.jsonl.gz"
|
||||
)
|
||||
kespeech_2_cuts = load_manifest_lazy(
|
||||
self.fbank_dir / "kespeech" / "kespeech-asr_cuts_train_phase2.jsonl.gz"
|
||||
)
|
||||
|
||||
return CutSet.mux(
|
||||
thchs_30_cuts,
|
||||
aishell_cuts,
|
||||
aishell_2_cuts,
|
||||
aishell_4_L_cuts,
|
||||
aishell_4_M_cuts,
|
||||
aishell_4_S_cuts,
|
||||
alimeeting_cuts,
|
||||
stcmds_cuts,
|
||||
primewords_cuts,
|
||||
magicdata_cuts,
|
||||
wenetspeech_L_cuts,
|
||||
kespeech_1_cuts,
|
||||
kespeech_2_cuts,
|
||||
weights=[
|
||||
len(thchs_30_cuts),
|
||||
len(aishell_cuts),
|
||||
len(aishell_2_cuts),
|
||||
len(aishell_4_L_cuts),
|
||||
len(aishell_4_M_cuts),
|
||||
len(aishell_4_S_cuts),
|
||||
len(alimeeting_cuts),
|
||||
len(stcmds_cuts),
|
||||
len(primewords_cuts),
|
||||
len(magicdata_cuts),
|
||||
len(wenetspeech_L_cuts),
|
||||
len(kespeech_1_cuts),
|
||||
len(kespeech_2_cuts),
|
||||
],
|
||||
)
|
||||
|
||||
def dev_cuts(self) -> CutSet:
|
||||
logging.info("About to get multidataset dev cuts")
|
||||
|
||||
# WeNetSpeech
|
||||
logging.info("Loading WeNetSpeech DEV set in lazy mode")
|
||||
wenetspeech_dev_cuts = load_manifest_lazy(
|
||||
self.fbank_dir / "wenetspeech" / "cuts_DEV_fixed.jsonl.gz"
|
||||
)
|
||||
|
||||
return wenetspeech_dev_cuts
|
||||
|
||||
def test_cuts(self) -> Dict[str, CutSet]:
|
||||
logging.info("About to get multidataset test cuts")
|
||||
|
||||
# AISHELL
|
||||
logging.info("Loading Aishell set in lazy mode")
|
||||
aishell_test_cuts = load_manifest_lazy(
|
||||
self.fbank_dir / "aishell_cuts_test.jsonl.gz"
|
||||
)
|
||||
aishell_dev_cuts = load_manifest_lazy(
|
||||
self.fbank_dir / "aishell_cuts_dev.jsonl.gz"
|
||||
)
|
||||
|
||||
# AISHELL-2
|
||||
logging.info("Loading Aishell-2 set in lazy mode")
|
||||
aishell2_test_cuts = load_manifest_lazy(
|
||||
self.fbank_dir / "aishell2_cuts_test.jsonl.gz"
|
||||
)
|
||||
aishell2_dev_cuts = load_manifest_lazy(
|
||||
self.fbank_dir / "aishell2_cuts_dev.jsonl.gz"
|
||||
)
|
||||
|
||||
# AISHELL-4
|
||||
logging.info("Loading Aishell-4 TEST set in lazy mode")
|
||||
aishell4_test_cuts = load_manifest_lazy(
|
||||
self.fbank_dir / "aishell4_cuts_test.jsonl.gz"
|
||||
)
|
||||
|
||||
# Ali-Meeting
|
||||
logging.info("Loading Ali-Meeting set in lazy mode")
|
||||
alimeeting_test_cuts = load_manifest_lazy(
|
||||
self.fbank_dir / "alimeeting-far_cuts_test.jsonl.gz"
|
||||
)
|
||||
alimeeting_eval_cuts = load_manifest_lazy(
|
||||
self.fbank_dir / "alimeeting-far_cuts_eval.jsonl.gz"
|
||||
)
|
||||
|
||||
# MagicData
|
||||
logging.info("Loading MagicData set in lazy mode")
|
||||
magicdata_test_cuts = load_manifest_lazy(
|
||||
self.fbank_dir / "magicdata_cuts_test.jsonl.gz"
|
||||
)
|
||||
magicdata_dev_cuts = load_manifest_lazy(
|
||||
self.fbank_dir / "magicdata_cuts_dev.jsonl.gz"
|
||||
)
|
||||
|
||||
# KeSpeech
|
||||
logging.info("Loading KeSpeech set in lazy mode")
|
||||
kespeech_test_cuts = load_manifest_lazy(
|
||||
self.fbank_dir / "kespeech" / "kespeech-asr_cuts_test.jsonl.gz"
|
||||
)
|
||||
kespeech_dev_phase1_cuts = load_manifest_lazy(
|
||||
self.fbank_dir / "kespeech" / "kespeech-asr_cuts_dev_phase1.jsonl.gz"
|
||||
)
|
||||
kespeech_dev_phase2_cuts = load_manifest_lazy(
|
||||
self.fbank_dir / "kespeech" / "kespeech-asr_cuts_dev_phase2.jsonl.gz"
|
||||
)
|
||||
|
||||
# WeNetSpeech
|
||||
logging.info("Loading WeNetSpeech set in lazy mode")
|
||||
wenetspeech_test_meeting_cuts = load_manifest_lazy(
|
||||
self.fbank_dir / "wenetspeech" / "cuts_TEST_MEETING.jsonl.gz"
|
||||
)
|
||||
wenetspeech_test_net_cuts = load_manifest_lazy(
|
||||
self.fbank_dir / "wenetspeech" / "cuts_TEST_NET.jsonl.gz"
|
||||
)
|
||||
wenetspeech_dev_cuts = load_manifest_lazy(
|
||||
self.fbank_dir / "wenetspeech" / "cuts_DEV_fixed.jsonl.gz"
|
||||
)
|
||||
|
||||
return {
|
||||
"wenetspeech-meeting_test": wenetspeech_test_meeting_cuts,
|
||||
# "aishell_test": aishell_test_cuts,
|
||||
# "aishell_dev": aishell_dev_cuts,
|
||||
# "ali-meeting_test": alimeeting_test_cuts,
|
||||
# "ali-meeting_eval": alimeeting_eval_cuts,
|
||||
# "aishell-4_test": aishell4_test_cuts,
|
||||
# "aishell-2_test": aishell2_test_cuts,
|
||||
# "aishell-2_dev": aishell2_dev_cuts,
|
||||
# "magicdata_test": magicdata_test_cuts,
|
||||
# "magicdata_dev": magicdata_dev_cuts,
|
||||
# "kespeech-asr_test": kespeech_test_cuts,
|
||||
# "kespeech-asr_dev_phase1": kespeech_dev_phase1_cuts,
|
||||
# "kespeech-asr_dev_phase2": kespeech_dev_phase2_cuts,
|
||||
# "wenetspeech-net_test": wenetspeech_test_net_cuts,
|
||||
# "wenetspeech_dev": wenetspeech_dev_cuts,
|
||||
}
|
||||
1
egs/multi_zh-hans/ASR/whisper/multi_dataset.py
Symbolic link
1
egs/multi_zh-hans/ASR/whisper/multi_dataset.py
Symbolic link
@ -0,0 +1 @@
|
||||
../../../speech_llm/ASR_LLM/whisper_llm_zh/multi_dataset.py
|
||||
@ -46,7 +46,7 @@ import torch.nn as nn
|
||||
from asr_datamodule import AsrDataModule
|
||||
from lhotse.cut import Cut
|
||||
from multi_dataset import MultiDataset
|
||||
from train import add_model_arguments, get_model, get_params
|
||||
from train import add_model_arguments, get_model, get_params, normalize_text_alimeeting
|
||||
|
||||
from icefall.checkpoint import (
|
||||
average_checkpoints,
|
||||
@ -367,19 +367,17 @@ def decode_dataset(
|
||||
hyps_dict = decode_one_batch(
|
||||
params=params,
|
||||
model=model,
|
||||
HLG=HLG,
|
||||
H=H,
|
||||
bpe_model=bpe_model,
|
||||
batch=batch,
|
||||
word_table=word_table,
|
||||
G=G,
|
||||
)
|
||||
|
||||
for name, hyps in hyps_dict.items():
|
||||
this_batch = []
|
||||
assert len(hyps) == len(texts)
|
||||
for cut_id, hyp_words, ref_text in zip(cut_ids, hyps, texts):
|
||||
ref_words = list(ref_text.replace(" ", ""))
|
||||
ref_text = normalize_text_alimeeting(ref_text)
|
||||
ref_words = ref_text.split()
|
||||
hyp_words = list("".join(hyp_words))
|
||||
this_batch.append((cut_id, ref_words, hyp_words))
|
||||
|
||||
@ -583,7 +581,7 @@ def main():
|
||||
data_module = AsrDataModule(args)
|
||||
multi_dataset = MultiDataset(args.manifest_dir)
|
||||
|
||||
test_sets_cuts = multi_dataset.test_cuts()
|
||||
test_sets_cuts = {**multi_dataset.test_cuts(), **multi_dataset.speechio_test_cuts()}
|
||||
|
||||
def remove_short_utt(c: Cut):
|
||||
T = ((c.num_frames - 7) // 2 + 1) // 2
|
||||
|
||||
@ -118,7 +118,7 @@ from beam_search import (
|
||||
)
|
||||
from lhotse.cut import Cut
|
||||
from multi_dataset import MultiDataset
|
||||
from train import add_model_arguments, get_model, get_params
|
||||
from train import add_model_arguments, get_model, get_params, normalize_text_alimeeting
|
||||
|
||||
from icefall.checkpoint import (
|
||||
average_checkpoints,
|
||||
@ -548,6 +548,8 @@ def decode_dataset(
|
||||
this_batch = []
|
||||
assert len(hyps) == len(texts)
|
||||
for cut_id, hyp_words, ref_text in zip(cut_ids, hyps, texts):
|
||||
ref_text = normalize_text_alimeeting(ref_text)
|
||||
ref_words = ref_text.split()
|
||||
hyp_text = "".join(hyp_words)
|
||||
this_batch.append((cut_id, ref_text, hyp_text))
|
||||
|
||||
@ -795,7 +797,7 @@ def main():
|
||||
)
|
||||
return T > 0
|
||||
|
||||
test_sets_cuts = multi_dataset.test_cuts()
|
||||
test_sets_cuts = {**multi_dataset.test_cuts(), **multi_dataset.speechio_test_cuts()}
|
||||
|
||||
test_sets = test_sets_cuts.keys()
|
||||
test_dl = [
|
||||
|
||||
@ -1,316 +0,0 @@
|
||||
# Copyright 2023 Xiaomi Corp. (authors: Zengrui Jin)
|
||||
#
|
||||
# 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 glob
|
||||
import logging
|
||||
import re
|
||||
from pathlib import Path
|
||||
from typing import Dict, List
|
||||
|
||||
import lhotse
|
||||
from lhotse import CutSet, load_manifest_lazy
|
||||
|
||||
|
||||
class MultiDataset:
|
||||
def __init__(self, fbank_dir: str):
|
||||
"""
|
||||
Args:
|
||||
manifest_dir:
|
||||
It is expected to contain the following files:
|
||||
- aidatatang_cuts_train.jsonl.gz
|
||||
- aishell_cuts_train.jsonl.gz
|
||||
- aishell2_cuts_train.jsonl.gz
|
||||
- aishell4_cuts_train_L.jsonl.gz
|
||||
- aishell4_cuts_train_M.jsonl.gz
|
||||
- aishell4_cuts_train_S.jsonl.gz
|
||||
- alimeeting-far_cuts_train.jsonl.gz
|
||||
- magicdata_cuts_train.jsonl.gz
|
||||
- primewords_cuts_train.jsonl.gz
|
||||
- stcmds_cuts_train.jsonl.gz
|
||||
- thchs_30_cuts_train.jsonl.gz
|
||||
- kespeech/kespeech-asr_cuts_train_phase1.jsonl.gz
|
||||
- kespeech/kespeech-asr_cuts_train_phase2.jsonl.gz
|
||||
- wenetspeech/cuts_L.jsonl.gz
|
||||
"""
|
||||
self.fbank_dir = Path(fbank_dir)
|
||||
|
||||
def train_cuts(self) -> CutSet:
|
||||
logging.info("About to get multidataset train cuts")
|
||||
|
||||
# THCHS-30
|
||||
logging.info("Loading THCHS-30 in lazy mode")
|
||||
thchs_30_cuts = load_manifest_lazy(
|
||||
self.fbank_dir / "thchs_30_cuts_train.jsonl.gz"
|
||||
)
|
||||
|
||||
# AISHELL-1
|
||||
logging.info("Loading Aishell-1 in lazy mode")
|
||||
aishell_cuts = load_manifest_lazy(
|
||||
self.fbank_dir / "aishell_cuts_train.jsonl.gz"
|
||||
)
|
||||
|
||||
# AISHELL-2
|
||||
logging.info("Loading Aishell-2 in lazy mode")
|
||||
aishell_2_cuts = load_manifest_lazy(
|
||||
self.fbank_dir / "aishell2_cuts_train.jsonl.gz"
|
||||
)
|
||||
|
||||
# AISHELL-4
|
||||
logging.info("Loading Aishell-4 in lazy mode")
|
||||
aishell_4_L_cuts = load_manifest_lazy(
|
||||
self.fbank_dir / "aishell4_cuts_train_L.jsonl.gz"
|
||||
)
|
||||
aishell_4_M_cuts = load_manifest_lazy(
|
||||
self.fbank_dir / "aishell4_cuts_train_M.jsonl.gz"
|
||||
)
|
||||
aishell_4_S_cuts = load_manifest_lazy(
|
||||
self.fbank_dir / "aishell4_cuts_train_S.jsonl.gz"
|
||||
)
|
||||
|
||||
# ST-CMDS
|
||||
logging.info("Loading ST-CMDS in lazy mode")
|
||||
stcmds_cuts = load_manifest_lazy(self.fbank_dir / "stcmds_cuts_train.jsonl.gz")
|
||||
|
||||
# Primewords
|
||||
logging.info("Loading Primewords in lazy mode")
|
||||
primewords_cuts = load_manifest_lazy(
|
||||
self.fbank_dir / "primewords_cuts_train.jsonl.gz"
|
||||
)
|
||||
|
||||
# MagicData
|
||||
logging.info("Loading MagicData in lazy mode")
|
||||
magicdata_cuts = load_manifest_lazy(
|
||||
self.fbank_dir / "magicdata_cuts_train.jsonl.gz"
|
||||
)
|
||||
|
||||
# Aidatatang_200zh
|
||||
logging.info("Loading Aidatatang_200zh in lazy mode")
|
||||
aidatatang_200zh_cuts = load_manifest_lazy(
|
||||
self.fbank_dir / "aidatatang_cuts_train.jsonl.gz"
|
||||
)
|
||||
|
||||
# Ali-Meeting
|
||||
logging.info("Loading Ali-Meeting in lazy mode")
|
||||
alimeeting_cuts = load_manifest_lazy(
|
||||
self.fbank_dir / "alimeeting-far_cuts_train.jsonl.gz"
|
||||
)
|
||||
|
||||
# WeNetSpeech
|
||||
logging.info("Loading WeNetSpeech in lazy mode")
|
||||
wenetspeech_L_cuts = load_manifest_lazy(
|
||||
self.fbank_dir / "wenetspeech" / "cuts_L.jsonl.gz"
|
||||
)
|
||||
|
||||
# KeSpeech
|
||||
logging.info("Loading KeSpeech in lazy mode")
|
||||
kespeech_1_cuts = load_manifest_lazy(
|
||||
self.fbank_dir / "kespeech" / "kespeech-asr_cuts_train_phase1.jsonl.gz"
|
||||
)
|
||||
kespeech_2_cuts = load_manifest_lazy(
|
||||
self.fbank_dir / "kespeech" / "kespeech-asr_cuts_train_phase2.jsonl.gz"
|
||||
)
|
||||
|
||||
return CutSet.mux(
|
||||
thchs_30_cuts,
|
||||
aishell_cuts,
|
||||
aishell_2_cuts,
|
||||
aishell_4_L_cuts,
|
||||
aishell_4_M_cuts,
|
||||
aishell_4_S_cuts,
|
||||
stcmds_cuts,
|
||||
primewords_cuts,
|
||||
magicdata_cuts,
|
||||
aidatatang_200zh_cuts,
|
||||
alimeeting_cuts,
|
||||
wenetspeech_L_cuts,
|
||||
kespeech_1_cuts,
|
||||
kespeech_2_cuts,
|
||||
weights=[
|
||||
len(thchs_30_cuts),
|
||||
len(aishell_cuts),
|
||||
len(aishell_2_cuts),
|
||||
len(aishell_4_L_cuts),
|
||||
len(aishell_4_M_cuts),
|
||||
len(aishell_4_S_cuts),
|
||||
len(stcmds_cuts),
|
||||
len(primewords_cuts),
|
||||
len(magicdata_cuts),
|
||||
len(aidatatang_200zh_cuts),
|
||||
len(alimeeting_cuts),
|
||||
len(wenetspeech_L_cuts),
|
||||
len(kespeech_1_cuts),
|
||||
len(kespeech_2_cuts),
|
||||
],
|
||||
)
|
||||
|
||||
def dev_cuts(self) -> CutSet:
|
||||
logging.info("About to get multidataset dev cuts")
|
||||
|
||||
# Aidatatang_200zh
|
||||
logging.info("Loading Aidatatang_200zh DEV set in lazy mode")
|
||||
aidatatang_dev_cuts = load_manifest_lazy(
|
||||
self.fbank_dir / "aidatatang_cuts_dev.jsonl.gz"
|
||||
)
|
||||
|
||||
# AISHELL
|
||||
logging.info("Loading Aishell DEV set in lazy mode")
|
||||
aishell_dev_cuts = load_manifest_lazy(
|
||||
self.fbank_dir / "aishell_cuts_dev.jsonl.gz"
|
||||
)
|
||||
|
||||
# AISHELL-2
|
||||
logging.info("Loading Aishell-2 DEV set in lazy mode")
|
||||
aishell2_dev_cuts = load_manifest_lazy(
|
||||
self.fbank_dir / "aishell2_cuts_dev.jsonl.gz"
|
||||
)
|
||||
|
||||
# Ali-Meeting
|
||||
logging.info("Loading Ali-Meeting DEV set in lazy mode")
|
||||
alimeeting_dev_cuts = load_manifest_lazy(
|
||||
self.fbank_dir / "alimeeting-far_cuts_eval.jsonl.gz"
|
||||
)
|
||||
|
||||
# MagicData
|
||||
logging.info("Loading MagicData DEV set in lazy mode")
|
||||
magicdata_dev_cuts = load_manifest_lazy(
|
||||
self.fbank_dir / "magicdata_cuts_dev.jsonl.gz"
|
||||
)
|
||||
|
||||
# KeSpeech
|
||||
logging.info("Loading KeSpeech DEV set in lazy mode")
|
||||
kespeech_dev_phase1_cuts = load_manifest_lazy(
|
||||
self.fbank_dir / "kespeech" / "kespeech-asr_cuts_dev_phase1.jsonl.gz"
|
||||
)
|
||||
kespeech_dev_phase2_cuts = load_manifest_lazy(
|
||||
self.fbank_dir / "kespeech" / "kespeech-asr_cuts_dev_phase2.jsonl.gz"
|
||||
)
|
||||
|
||||
# WeNetSpeech
|
||||
logging.info("Loading WeNetSpeech DEV set in lazy mode")
|
||||
wenetspeech_dev_cuts = load_manifest_lazy(
|
||||
self.fbank_dir / "wenetspeech" / "cuts_DEV.jsonl.gz"
|
||||
)
|
||||
|
||||
return wenetspeech_dev_cuts
|
||||
# return [
|
||||
# aidatatang_dev_cuts,
|
||||
# aishell_dev_cuts,
|
||||
# aishell2_dev_cuts,
|
||||
# alimeeting_dev_cuts,
|
||||
# magicdata_dev_cuts,
|
||||
# kespeech_dev_phase1_cuts,
|
||||
# kespeech_dev_phase2_cuts,
|
||||
# wenetspeech_dev_cuts,
|
||||
# ]
|
||||
|
||||
def test_cuts(self) -> Dict[str, CutSet]:
|
||||
logging.info("About to get multidataset test cuts")
|
||||
|
||||
# Aidatatang_200zh
|
||||
logging.info("Loading Aidatatang_200zh set in lazy mode")
|
||||
aidatatang_test_cuts = load_manifest_lazy(
|
||||
self.fbank_dir / "aidatatang_cuts_test.jsonl.gz"
|
||||
)
|
||||
aidatatang_dev_cuts = load_manifest_lazy(
|
||||
self.fbank_dir / "aidatatang_cuts_dev.jsonl.gz"
|
||||
)
|
||||
|
||||
# AISHELL
|
||||
logging.info("Loading Aishell set in lazy mode")
|
||||
aishell_test_cuts = load_manifest_lazy(
|
||||
self.fbank_dir / "aishell_cuts_test.jsonl.gz"
|
||||
)
|
||||
aishell_dev_cuts = load_manifest_lazy(
|
||||
self.fbank_dir / "aishell_cuts_dev.jsonl.gz"
|
||||
)
|
||||
|
||||
# AISHELL-2
|
||||
logging.info("Loading Aishell-2 set in lazy mode")
|
||||
aishell2_test_cuts = load_manifest_lazy(
|
||||
self.fbank_dir / "aishell2_cuts_test.jsonl.gz"
|
||||
)
|
||||
aishell2_dev_cuts = load_manifest_lazy(
|
||||
self.fbank_dir / "aishell2_cuts_dev.jsonl.gz"
|
||||
)
|
||||
|
||||
# AISHELL-4
|
||||
logging.info("Loading Aishell-4 TEST set in lazy mode")
|
||||
aishell4_test_cuts = load_manifest_lazy(
|
||||
self.fbank_dir / "aishell4_cuts_test.jsonl.gz"
|
||||
)
|
||||
|
||||
# Ali-Meeting
|
||||
logging.info("Loading Ali-Meeting set in lazy mode")
|
||||
alimeeting_test_cuts = load_manifest_lazy(
|
||||
self.fbank_dir / "alimeeting-far_cuts_test.jsonl.gz"
|
||||
)
|
||||
alimeeting_eval_cuts = load_manifest_lazy(
|
||||
self.fbank_dir / "alimeeting-far_cuts_eval.jsonl.gz"
|
||||
)
|
||||
|
||||
# MagicData
|
||||
logging.info("Loading MagicData set in lazy mode")
|
||||
magicdata_test_cuts = load_manifest_lazy(
|
||||
self.fbank_dir / "magicdata_cuts_test.jsonl.gz"
|
||||
)
|
||||
magicdata_dev_cuts = load_manifest_lazy(
|
||||
self.fbank_dir / "magicdata_cuts_dev.jsonl.gz"
|
||||
)
|
||||
|
||||
# KeSpeech
|
||||
logging.info("Loading KeSpeech set in lazy mode")
|
||||
kespeech_test_cuts = load_manifest_lazy(
|
||||
self.fbank_dir / "kespeech" / "kespeech-asr_cuts_test.jsonl.gz"
|
||||
)
|
||||
kespeech_dev_phase1_cuts = load_manifest_lazy(
|
||||
self.fbank_dir / "kespeech" / "kespeech-asr_cuts_dev_phase1.jsonl.gz"
|
||||
)
|
||||
kespeech_dev_phase2_cuts = load_manifest_lazy(
|
||||
self.fbank_dir / "kespeech" / "kespeech-asr_cuts_dev_phase2.jsonl.gz"
|
||||
)
|
||||
|
||||
# WeNetSpeech
|
||||
logging.info("Loading WeNetSpeech set in lazy mode")
|
||||
wenetspeech_test_meeting_cuts = load_manifest_lazy(
|
||||
self.fbank_dir / "wenetspeech" / "cuts_TEST_MEETING.jsonl.gz"
|
||||
)
|
||||
wenetspeech_test_net_cuts = load_manifest_lazy(
|
||||
self.fbank_dir / "wenetspeech" / "cuts_TEST_NET.jsonl.gz"
|
||||
)
|
||||
wenetspeech_dev_cuts = load_manifest_lazy(
|
||||
self.fbank_dir / "wenetspeech" / "cuts_DEV.jsonl.gz"
|
||||
)
|
||||
|
||||
return {
|
||||
"aidatatang_test": aidatatang_test_cuts,
|
||||
"aidatatang_dev": aidatatang_dev_cuts,
|
||||
"alimeeting_test": alimeeting_test_cuts,
|
||||
"alimeeting_eval": alimeeting_eval_cuts,
|
||||
"aishell_test": aishell_test_cuts,
|
||||
"aishell_dev": aishell_dev_cuts,
|
||||
"aishell-2_test": aishell2_test_cuts,
|
||||
"aishell-2_dev": aishell2_dev_cuts,
|
||||
"aishell-4": aishell4_test_cuts,
|
||||
"magicdata_test": magicdata_test_cuts,
|
||||
"magicdata_dev": magicdata_dev_cuts,
|
||||
"kespeech-asr_test": kespeech_test_cuts,
|
||||
"kespeech-asr_dev_phase1": kespeech_dev_phase1_cuts,
|
||||
"kespeech-asr_dev_phase2": kespeech_dev_phase2_cuts,
|
||||
"wenetspeech-meeting_test": wenetspeech_test_meeting_cuts,
|
||||
"wenetspeech-net_test": wenetspeech_test_net_cuts,
|
||||
"wenetspeech_dev": wenetspeech_dev_cuts,
|
||||
}
|
||||
1
egs/multi_zh-hans/ASR/zipformer/multi_dataset.py
Symbolic link
1
egs/multi_zh-hans/ASR/zipformer/multi_dataset.py
Symbolic link
@ -0,0 +1 @@
|
||||
../../../speech_llm/ASR_LLM/whisper_llm_zh/multi_dataset.py
|
||||
@ -539,6 +539,43 @@ def get_params() -> AttributeDict:
|
||||
return params
|
||||
|
||||
|
||||
def normalize_text_alimeeting(text: str, normalize: str = "m2met") -> str:
|
||||
"""
|
||||
Text normalization similar to M2MeT challenge baseline.
|
||||
See: https://github.com/yufan-aslp/AliMeeting/blob/main/asr/local/text_normalize.pl
|
||||
"""
|
||||
if normalize == "none":
|
||||
return text
|
||||
elif normalize == "m2met":
|
||||
import re
|
||||
|
||||
text = text.replace(" ", "")
|
||||
text = text.replace("<sil>", "")
|
||||
text = text.replace("<%>", "")
|
||||
text = text.replace("<->", "")
|
||||
text = text.replace("<$>", "")
|
||||
text = text.replace("<#>", "")
|
||||
text = text.replace("<_>", "")
|
||||
text = text.replace("<space>", "")
|
||||
text = text.replace("`", "")
|
||||
text = text.replace("&", "")
|
||||
text = text.replace(",", "")
|
||||
if re.search("[a-zA-Z]", text):
|
||||
text = text.upper()
|
||||
text = text.replace("Ａ", "A")
|
||||
text = text.replace("ａ", "A")
|
||||
text = text.replace("ｂ", "B")
|
||||
text = text.replace("ｃ", "C")
|
||||
text = text.replace("ｋ", "K")
|
||||
text = text.replace("ｔ", "T")
|
||||
text = text.replace("，", "")
|
||||
text = text.replace("丶", "")
|
||||
text = text.replace("。", "")
|
||||
text = text.replace("、", "")
|
||||
text = text.replace("？", "")
|
||||
return text
|
||||
|
||||
|
||||
def _to_int_tuple(s: str):
|
||||
return tuple(map(int, s.split(",")))
|
||||
|
||||
@ -775,42 +812,6 @@ def compute_loss(
|
||||
warmup: a floating point value which increases throughout training;
|
||||
values >= 1.0 are fully warmed up and have all modules present.
|
||||
"""
|
||||
def normalize_text_alimeeting(text: str, normalize: str = "m2met") -> str:
|
||||
"""
|
||||
Text normalization similar to M2MeT challenge baseline.
|
||||
See: https://github.com/yufan-aslp/AliMeeting/blob/main/asr/local/text_normalize.pl
|
||||
"""
|
||||
if normalize == "none":
|
||||
return text
|
||||
elif normalize == "m2met":
|
||||
import re
|
||||
|
||||
text = text.replace(" ", "")
|
||||
text = text.replace("<sil>", "")
|
||||
text = text.replace("<%>", "")
|
||||
text = text.replace("<->", "")
|
||||
text = text.replace("<$>", "")
|
||||
text = text.replace("<#>", "")
|
||||
text = text.replace("<_>", "")
|
||||
text = text.replace("<space>", "")
|
||||
text = text.replace("`", "")
|
||||
text = text.replace("&", "")
|
||||
text = text.replace(",", "")
|
||||
if re.search("[a-zA-Z]", text):
|
||||
text = text.upper()
|
||||
text = text.replace("Ａ", "A")
|
||||
text = text.replace("ａ", "A")
|
||||
text = text.replace("ｂ", "B")
|
||||
text = text.replace("ｃ", "C")
|
||||
text = text.replace("ｋ", "K")
|
||||
text = text.replace("ｔ", "T")
|
||||
text = text.replace("，", "")
|
||||
text = text.replace("丶", "")
|
||||
text = text.replace("。", "")
|
||||
text = text.replace("、", "")
|
||||
text = text.replace("？", "")
|
||||
return text
|
||||
|
||||
device = model.device if isinstance(model, DDP) else next(model.parameters()).device
|
||||
feature = batch["inputs"]
|
||||
# at entry, feature is (N, T, C)
|
||||
|
||||
Loading…
x
Reference in New Issue
Block a user