mirrors/icefall
1
0
Fork 0
mirror of https://github.com/k2-fsa/icefall.git synced 2025-08-12 19:42:19 +00:00
Code Issues Packages Projects Releases Wiki Activity

Fix style and add copyright

Browse Source
This commit is contained in:
pkufool 2021-08-22 22:06:28 +08:00
parent 8469f9ae0a
commit b4fd6338bb
34 changed files with 646 additions and 103 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

3
.flake8
View File

@ -2,6 +2,9 @@
show-source=true
statistics=true
max-line-length = 80
per-file-ignores =
# line too long
egs/librispeech/ASR/conformer_ctc/conformer.py: E501,
exclude =
.git,

196
egs/librispeech/ASR/conformer_ctc/conformer.py
View File

@ -1,7 +1,21 @@
#!/usr/bin/env python3
# Copyright 2021 Xiaomi Corp. (authors: Fangjun Kuang)
# 2021 University of Chinese Academy of Sciences (author: Han Zhu)
#
# 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.
# Copyright (c) 2021 University of Chinese Academy of Sciences (author: Han Zhu)
# Apache 2.0
import math
import warnings
@ -15,19 +29,19 @@ from transformer import Supervisions, Transformer, encoder_padding_mask
class Conformer(Transformer):
"""
Args:
num_features (int): Number of input features
num_classes (int): Number of output classes
subsampling_factor (int): subsampling factor of encoder (the convolution layers before transformers)
d_model (int): attention dimension
nhead (int): number of head
dim_feedforward (int): feedforward dimention
num_encoder_layers (int): number of encoder layers
num_decoder_layers (int): number of decoder layers
dropout (float): dropout rate
cnn_module_kernel (int): Kernel size of convolution module
normalize_before (bool): whether to use layer_norm before the first block.
vgg_frontend (bool): whether to use vgg frontend.
"""
num_features (int): Number of input features
num_classes (int): Number of output classes
subsampling_factor (int): subsampling factor of encoder (the convolution layers before transformers)
d_model (int): attention dimension
nhead (int): number of head
dim_feedforward (int): feedforward dimention
num_encoder_layers (int): number of encoder layers
num_decoder_layers (int): number of decoder layers
dropout (float): dropout rate
cnn_module_kernel (int): Kernel size of convolution module
normalize_before (bool): whether to use layer_norm before the first block.
vgg_frontend (bool): whether to use vgg frontend.
""" # noqa E501
def __init__(
self,
@ -80,8 +94,8 @@ class Conformer(Transformer):
if self.normalize_before and self.is_espnet_structure:
self.after_norm = nn.LayerNorm(d_model)
else:
# Note: TorchScript detects that self.after_norm could be used inside forward()
# and throws an error without this change.
# Note: TorchScript detects that self.after_norm could be used
# inside forward() and throws an error without this change.
self.after_norm = identity
def run_encoder(
@ -93,7 +107,7 @@ class Conformer(Transformer):
The model input. Its shape is [N, T, C].
supervisions:
Supervision in lhotse format.
See https://github.com/lhotse-speech/lhotse/blob/master/lhotse/dataset/speech_recognition.py#L32 # noqa
See https://github.com/lhotse-speech/lhotse/blob/master/lhotse/dataset/speech_recognition.py#L32
CAUTION: It contains length information, i.e., start and number of
frames, before subsampling
It is read directly from the batch, without any sorting. It is used
@ -119,17 +133,18 @@ class Conformer(Transformer):
class ConformerEncoderLayer(nn.Module):
"""
ConformerEncoderLayer is made up of self-attn, feedforward and convolution networks.
"""ConformerEncoderLayer is made up of self-attn, feedforward and
convolution networks.
See: "Conformer: Convolution-augmented Transformer for Speech Recognition"
Args:
d_model: the number of expected features in the input (required).
nhead: the number of heads in the multiheadattention models (required).
dim_feedforward: the dimension of the feedforward network model (default=2048).
dropout: the dropout value (default=0.1).
cnn_module_kernel (int): Kernel size of convolution module.
normalize_before: whether to use layer_norm before the first block.
d_model: the number of expected features in the input (required).
nhead: the number of heads in the multiheadattention models (required).
dim_feedforward: the dimension of the feedforward network model (default=2048).
dropout: the dropout value (default=0.1).
cnn_module_kernel (int): Kernel size of convolution module.
normalize_before: whether to use layer_norm before the first block.
Examples::
>>> encoder_layer = ConformerEncoderLayer(d_model=512, nhead=8)
@ -193,8 +208,7 @@ class ConformerEncoderLayer(nn.Module):
src_mask: Optional[Tensor] = None,
src_key_padding_mask: Optional[Tensor] = None,
) -> Tensor:
"""
Pass the input through the encoder layer.
"""Pass the input through the encoder layer.
Args:
src: the sequence to the encoder layer (required).
@ -301,7 +315,8 @@ class ConformerEncoder(nn.TransformerEncoder):
pos_emb: (N, 2*S-1, E)
mask: (S, S).
src_key_padding_mask: (N, S).
S is the source sequence length, T is the target sequence length, N is the batch size, E is the feature number
S is the source sequence length, T is the target sequence length,
N is the batch size, E is the feature number
"""
output = src
@ -396,7 +411,7 @@ class RelPositionalEncoding(torch.nn.Module):
:,
self.pe.size(1) // 2
- x.size(1)
+ 1 : self.pe.size(1) // 2
+ 1: self.pe.size(1) // 2
+ x.size(1),
]
return self.dropout(x), self.dropout(pos_emb)
@ -470,42 +485,48 @@ class RelPositionMultiheadAttention(nn.Module):
Args:
query, key, value: map a query and a set of key-value pairs to an output.
pos_emb: Positional embedding tensor
key_padding_mask: if provided, specified padding elements in the key will
be ignored by the attention. When given a binary mask and a value is True,
the corresponding value on the attention layer will be ignored. When given
a byte mask and a value is non-zero, the corresponding value on the attention
layer will be ignored
key_padding_mask: if provided, specified padding elements in the key
will be ignored by the attention. When given a binary mask and
a value is True, the corresponding value on the attention layer
will be ignored. When given a byte mask and a value is non-zero,
the corresponding value on the attention layer will be ignored.
need_weights: output attn_output_weights.
attn_mask: 2D or 3D mask that prevents attention to certain positions. A 2D mask will be broadcasted for all
the batches while a 3D mask allows to specify a different mask for the entries of each batch.
attn_mask: 2D or 3D mask that prevents attention to certain positions.
A 2D mask will be broadcasted for all the batches while a 3D mask
allows to specify a different mask for the entries of each batch.
Shape:
- Inputs:
- query: :math:`(L, N, E)` where L is the target sequence length, N is the batch size, E is
the embedding dimension.
- key: :math:`(S, N, E)`, where S is the source sequence length, N is the batch size, E is
the embedding dimension.
- value: :math:`(S, N, E)` where S is the source sequence length, N is the batch size, E is
the embedding dimension.
- pos_emb: :math:`(N, 2*L-1, E)` where L is the target sequence length, N is the batch size, E is
the embedding dimension.
- key_padding_mask: :math:`(N, S)` where N is the batch size, S is the source sequence length.
If a ByteTensor is provided, the non-zero positions will be ignored while the position
with the zero positions will be unchanged. If a BoolTensor is provided, the positions with the
value of ``True`` will be ignored while the position with the value of ``False`` will be unchanged.
- attn_mask: 2D mask :math:`(L, S)` where L is the target sequence length, S is the source sequence length.
3D mask :math:`(N*num_heads, L, S)` where N is the batch size, L is the target sequence length,
S is the source sequence length. attn_mask ensure that position i is allowed to attend the unmasked
positions. If a ByteTensor is provided, the non-zero positions are not allowed to attend
while the zero positions will be unchanged. If a BoolTensor is provided, positions with ``True``
is not allowed to attend while ``False`` values will be unchanged. If a FloatTensor
is provided, it will be added to the attention weight.
- query: :math:`(L, N, E)` where L is the target sequence length,
N is the batch size, E is the embedding dimension.
- key: :math:`(S, N, E)`, where S is the source sequence length,
N is the batch size, E is the embedding dimension.
- value: :math:`(S, N, E)` where S is the source sequence length,
N is the batch size, E is the embedding dimension.
- pos_emb: :math:`(N, 2*L-1, E)` where L is the target sequence length,
N is the batch size, E is the embedding dimension.
- key_padding_mask: :math:`(N, S)` where N is the batch size,
S is the source sequence length. If a ByteTensor is provided,
the non-zero positions will be ignored while the position
with the zero positions will be unchanged. If a BoolTensor is provided,
the positions with the value of ``True`` will be ignored while
the position with the value of ``False`` will be unchanged.
- attn_mask: 2D mask :math:`(L, S)` where L is the target sequence length,
S is the source sequence length. 3D mask :math:`(N*num_heads, L, S)`
where N is the batch size, L is the target sequence length,
S is the source sequence length. attn_mask ensure that position
i is allowed to attend the unmasked positions.
If a ByteTensor is provided, the non-zero positions are not
allowed to attend while the zero positions will be unchanged.
If a BoolTensor is provided, positions with ``True`` is not allowed
to attend while ``False`` values will be unchanged.
If a FloatTensor is provided, it will be added to the attention weight.
- Outputs:
- attn_output: :math:`(L, N, E)` where L is the target sequence length, N is the batch size,
E is the embedding dimension.
- attn_output: :math:`(L, N, E)` where L is the target sequence length,
N is the batch size, E is the embedding dimension.
- attn_output_weights: :math:`(N, L, S)` where N is the batch size,
L is the target sequence length, S is the source sequence length.
L is the target sequence length, S is the source sequence length.
"""
return self.multi_head_attention_forward(
query,
@ -582,36 +603,43 @@ class RelPositionMultiheadAttention(nn.Module):
be ignored by the attention. This is an binary mask. When the value is True,
the corresponding value on the attention layer will be filled with -inf.
need_weights: output attn_output_weights.
attn_mask: 2D or 3D mask that prevents attention to certain positions. A 2D mask will be broadcasted for all
the batches while a 3D mask allows to specify a different mask for the entries of each batch.
attn_mask: 2D or 3D mask that prevents attention to certain positions.
A 2D mask will be broadcasted for all the batches while a 3D mask
allows to specify a different mask for the entries of each batch.
Shape:
Inputs:
- query: :math:`(L, N, E)` where L is the target sequence length, N is the batch size, E is
the embedding dimension.
- key: :math:`(S, N, E)`, where S is the source sequence length, N is the batch size, E is
the embedding dimension.
- value: :math:`(S, N, E)` where S is the source sequence length, N is the batch size, E is
the embedding dimension.
- pos_emb: :math:`(N, 2*L-1, E)` or :math:`(1, 2*L-1, E)` where L is the target sequence
length, N is the batch size, E is the embedding dimension.
- key_padding_mask: :math:`(N, S)` where N is the batch size, S is the source sequence length.
If a ByteTensor is provided, the non-zero positions will be ignored while the zero positions
will be unchanged. If a BoolTensor is provided, the positions with the
value of ``True`` will be ignored while the position with the value of ``False`` will be unchanged.
- attn_mask: 2D mask :math:`(L, S)` where L is the target sequence length, S is the source sequence length.
3D mask :math:`(N*num_heads, L, S)` where N is the batch size, L is the target sequence length,
S is the source sequence length. attn_mask ensures that position i is allowed to attend the unmasked
positions. If a ByteTensor is provided, the non-zero positions are not allowed to attend
while the zero positions will be unchanged. If a BoolTensor is provided, positions with ``True``
are not allowed to attend while ``False`` values will be unchanged. If a FloatTensor
is provided, it will be added to the attention weight.
- query: :math:`(L, N, E)` where L is the target sequence length,
N is the batch size, E is the embedding dimension.
- key: :math:`(S, N, E)`, where S is the source sequence length,
N is the batch size, E is the embedding dimension.
- value: :math:`(S, N, E)` where S is the source sequence length,
N is the batch size, E is the embedding dimension.
- pos_emb: :math:`(N, 2*L-1, E)` or :math:`(1, 2*L-1, E)` where
L is the target sequencelength, N is the batch size,
E is the embedding dimension.
- key_padding_mask: :math:`(N, S)` where N is the batch size,
S is the source sequence length. If a ByteTensor is provided,
the non-zero positions will be ignored while the zero positions
will be unchanged. If a BoolTensor is provided, the positions with the
value of ``True`` will be ignored while the position with the
value of ``False`` will be unchanged.
- attn_mask: 2D mask :math:`(L, S)` where L is the target sequence length,
S is the source sequence length. 3D mask :math:`(N*num_heads, L, S)`
where N is the batch size, L is the target sequence length,
S is the source sequence length. attn_mask ensures that position
i is allowed to attend the unmasked positions.
If a ByteTensor is provided, the non-zero positions are not
allowed to attend while the zero positions will be unchanged.
If a BoolTensor is provided, positions with ``True`` are not
allowed to attend while ``False`` values will be unchanged.
If a FloatTensor is provided, it will be added to the attention weight.
Outputs:
- attn_output: :math:`(L, N, E)` where L is the target sequence length, N is the batch size,
E is the embedding dimension.
- attn_output: :math:`(L, N, E)` where L is the target sequence length,
N is the batch size, E is the embedding dimension.
- attn_output_weights: :math:`(N, L, S)` where N is the batch size,
L is the target sequence length, S is the source sequence length.
L is the target sequence length, S is the source sequence length.
"""
tgt_len, bsz, embed_dim = query.size()
@ -844,7 +872,7 @@ class ConvolutionModule(nn.Module):
kernel_size (int): Kernerl size of conv layers.
bias (bool): Whether to use bias in conv layers (default=True).
"""
""" # noqa
def __init__(
self, channels: int, kernel_size: int, bias: bool = True

16
egs/librispeech/ASR/conformer_ctc/decode.py
View File

@ -1,8 +1,20 @@
#!/usr/bin/env python3
# Copyright 2021 Xiaomi Corporation (Author: Liyong Guo, 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.
# (still working in progress)
import argparse
import logging

23
egs/librispeech/ASR/conformer_ctc/pretrained.py
View File

@ -1,4 +1,20 @@
#!/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.
import argparse
import logging
@ -59,7 +75,7 @@ def get_parser():
decoding lattice and then use 1best to decode the
rescored lattice.
We call it HLG decoding + n-gram LM rescoring.
(3) attention-decoder - Extract n paths from he rescored
(3) attention-decoder - Extract n paths from the rescored
lattice and use the transformer attention decoder for
rescoring.
We call it HLG decoding + n-gram LM rescoring + attention
@ -248,6 +264,7 @@ def main():
G = G.to(device)
# Add epsilon self-loops to G as we will compose
# it with the whole lattice later
G = G.to(device)
G = k2.add_epsilon_self_loops(G)
G = k2.arc_sort(G)
G.lm_scores = G.scores.clone()
@ -268,7 +285,7 @@ def main():
)
waves = [w.to(device) for w in waves]
logging.info(f"Decoding started")
logging.info("Decoding started")
features = fbank(waves)
features = pad_sequence(
@ -338,7 +355,7 @@ def main():
s += f"{filename}:\n{words}\n\n"
logging.info(s)
logging.info(f"Decoding Done")
logging.info("Decoding Done")
if __name__ == "__main__":

17
egs/librispeech/ASR/conformer_ctc/subsampling.py
View File

@ -1,3 +1,20 @@
# 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.
import torch
import torch.nn as nn

16
egs/librispeech/ASR/conformer_ctc/test_subsampling.py
View File

@ -1,4 +1,20 @@
#!/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.
from subsampling import Conv2dSubsampling
from subsampling import VggSubsampling

16
egs/librispeech/ASR/conformer_ctc/test_transformer.py
View File

@ -1,4 +1,20 @@
#!/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.
import torch
from transformer import (

16
egs/librispeech/ASR/conformer_ctc/train.py
View File

@ -1,6 +1,20 @@
#!/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 is just at the very beginning ...
import argparse
import logging

24
egs/librispeech/ASR/conformer_ctc/transformer.py
View File

@ -1,5 +1,19 @@
# Copyright (c) 2021 University of Chinese Academy of Sciences (author: Han Zhu)
# Apache 2.0
# Copyright 2021 University of Chinese Academy of Sciences (author: Han Zhu)
#
# 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 math
from typing import Dict, List, Optional, Tuple
@ -779,7 +793,8 @@ class Noam(object):
class LabelSmoothingLoss(nn.Module):
"""
Label-smoothing loss. KL-divergence between q_{smoothed ground truth prob.}(w)
Label-smoothing loss. KL-divergence between
q_{smoothed ground truth prob.}(w)
and p_{prob. computed by model}(w) is minimized.
Modified from
https://github.com/espnet/espnet/blob/master/espnet/nets/pytorch_backend/transformer/label_smoothing_loss.py # noqa
@ -864,7 +879,8 @@ def encoder_padding_mask(
frames, before subsampling)
Returns:
Tensor: Mask tensor of dimension (batch_size, input_length), True denote the masked indices.
Tensor: Mask tensor of dimension (batch_size, input_length),
True denote the masked indices.
"""
if supervisions is None:
return None

16
egs/librispeech/ASR/local/compile_hlg.py
View File

@ -1,4 +1,20 @@
#!/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 script takes as input lang_dir and generates HLG from

21
egs/librispeech/ASR/local/compute_fbank_librispeech.py
View File

@ -1,4 +1,20 @@
#!/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 LibriSpeech dataset.
@ -17,8 +33,9 @@ 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
# 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)

21
egs/librispeech/ASR/local/compute_fbank_musan.py
View File

@ -1,4 +1,20 @@
#!/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 musan dataset.
@ -17,8 +33,9 @@ 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
# 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)

17
egs/librispeech/ASR/local/download_lm.py
View File

@ -1,6 +1,21 @@
#!/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.
# Copyright (c) 2021 Xiaomi Corporation (authors: Fangjun Kuang)
"""
This file downloads the following LibriSpeech LM files:

16
egs/librispeech/ASR/local/prepare_lang.py
View File

@ -1,6 +1,20 @@
#!/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.
# Copyright (c) 2021 Xiaomi Corporation (authors: Fangjun Kuang)
"""
This script takes as input a lexicon file "data/lang_phone/lexicon.txt"

16
egs/librispeech/ASR/local/prepare_lang_bpe.py
View File

@ -1,4 +1,20 @@
#!/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.
# Copyright (c) 2021 Xiaomi Corporation (authors: Fangjun Kuang)

16
egs/librispeech/ASR/local/test_prepare_lang.py
View File

@ -1,4 +1,20 @@
#!/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.
# Copyright (c) 2021 Xiaomi Corporation (authors: Fangjun Kuang)

16
egs/librispeech/ASR/local/train_bpe_model.py
View File

@ -1,4 +1,20 @@
#!/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.
# You can install sentencepiece via:
#

17
egs/librispeech/ASR/tdnn_lstm_ctc/asr_datamodule.py
View File

@ -1,3 +1,20 @@
# 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.
import argparse
import logging
from functools import lru_cache

15
egs/librispeech/ASR/tdnn_lstm_ctc/decode.py
View File

@ -1,4 +1,19 @@
#!/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.
import argparse

17
egs/librispeech/ASR/tdnn_lstm_ctc/model.py
View File

@ -1,3 +1,20 @@
# 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.
import torch
import torch.nn as nn

16
egs/librispeech/ASR/tdnn_lstm_ctc/train.py
View File

@ -1,4 +1,20 @@
#!/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.
import argparse
import logging

17
icefall/bpe_graph_compiler.py
View File

@ -1,3 +1,20 @@
# 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.
from pathlib import Path
from typing import List, Union

17
icefall/checkpoint.py
View File

@ -1,3 +1,20 @@
# 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.
import logging
from pathlib import Path
from typing import Any, Dict, List, Optional, Union

17
icefall/dataset/datamodule.py
View File

@ -1,3 +1,20 @@
# 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.
import argparse
from typing import List, Union

16
icefall/decode.py
View File

@ -1,3 +1,19 @@
# 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.
import logging
from typing import Dict, List, Optional, Tuple, Union

17
icefall/dist.py
View File

@ -1,3 +1,20 @@
# 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.
import os
import torch

17
icefall/graph_compiler.py
View File

@ -1,3 +1,20 @@
# 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.
from typing import List
import k2

17
icefall/lexicon.py
View File

@ -1,3 +1,20 @@
# 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.
import logging
import re
import sys

17
icefall/utils.py
View File

@ -1,3 +1,20 @@
# 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.
import argparse
import logging
import os

18
test/test_bpe_graph_compiler.py
View File

@ -1,6 +1,20 @@
#!/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.
# Copyright (c) 2021 Xiaomi Corporation (authors: Fangjun Kuang)
from icefall.bpe_graph_compiler import BpeCtcTrainingGraphCompiler
from icefall.lexicon import BpeLexicon
@ -15,7 +29,7 @@ def test():
compiler = BpeCtcTrainingGraphCompiler(lang_dir)
ids = compiler.texts_to_ids(["HELLO", "WORLD ZZZ"])
fsa = compiler.compile(ids)
compiler.compile(ids)
lexicon = BpeLexicon(lang_dir)
ids0 = lexicon.words_to_piece_ids(["HELLO"])

16
test/test_checkpoint.py
View File

@ -1,4 +1,20 @@
#!/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.
import pytest
import torch

16
test/test_graph_compiler.py
View File

@ -1,6 +1,20 @@
#!/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.
# Copyright (c) 2021 Xiaomi Corporation (authors: Fangjun Kuang)
import re

16
test/test_lexicon.py
View File

@ -1,4 +1,20 @@
#!/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.
from pathlib import Path

17
test/test_utils.py
View File

@ -1,4 +1,21 @@
#!/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.
import k2
import pytest
import torch