You can not select more than 25 topics Topics must start with a letter or number, can include dashes ('-') and can be up to 35 characters long.
PaddleSpeech/parakeet/modules/losses.py

126 lines
4.1 KiB

# Copyright (c) 2020 PaddlePaddle Authors. All Rights Reserved.
#
# 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 paddle
from paddle.fluid.layers import sequence_mask
from paddle.nn import functional as F
__all__ = [
"guided_attention_loss",
"weighted_mean",
"masked_l1_loss",
"masked_softmax_with_cross_entropy",
]
def attention_guide(dec_lens, enc_lens, N, T, g, dtype=None):
"""Build that W matrix. shape(B, T_dec, T_enc)
W[i, n, t] = 1 - exp(-(n/dec_lens[i] - t/enc_lens[i])**2 / (2g**2))
See also:
Tachibana, Hideyuki, Katsuya Uenoyama, and Shunsuke Aihara. 2017. “Efficiently Trainable Text-to-Speech System Based on Deep Convolutional Networks with Guided Attention.” ArXiv:1710.08969 [Cs, Eess], October. http://arxiv.org/abs/1710.08969.
"""
dtype = dtype or paddle.get_default_dtype()
dec_pos = paddle.arange(0, N).astype(dtype) / dec_lens.unsqueeze(
-1) # n/N # shape(B, T_dec)
enc_pos = paddle.arange(0, T).astype(dtype) / enc_lens.unsqueeze(
-1) # t/T # shape(B, T_enc)
W = 1 - paddle.exp(-(dec_pos.unsqueeze(-1) - enc_pos.unsqueeze(1))**2 /
(2 * g**2))
dec_mask = sequence_mask(dec_lens, maxlen=N)
enc_mask = sequence_mask(enc_lens, maxlen=T)
mask = dec_mask.unsqueeze(-1) * enc_mask.unsqueeze(1)
mask = paddle.cast(mask, W.dtype)
W *= mask
return W
def guided_attention_loss(attention_weight, dec_lens, enc_lens, g):
"""Guided attention loss, masked to excluded padding parts."""
_, N, T = attention_weight.shape
W = attention_guide(dec_lens, enc_lens, N, T, g, attention_weight.dtype)
total_tokens = (dec_lens * enc_lens).astype(W.dtype)
loss = paddle.mean(paddle.sum(W * attention_weight, [1, 2]) / total_tokens)
return loss
def weighted_mean(input, weight):
"""Weighted mean. It can also be used as masked mean.
Parameters
-----------
input : Tensor
The input tensor.
weight : Tensor
The weight tensor with broadcastable shape with the input.
Returns
----------
Tensor [shape=(1,)]
Weighted mean tensor with the same dtype as input.
"""
weight = paddle.cast(weight, input.dtype)
broadcast_ratio = input.size / weight.size
return paddle.sum(input * weight) / (paddle.sum(weight) * broadcast_ratio)
def masked_l1_loss(prediction, target, mask):
"""Compute maksed L1 loss.
Parameters
----------
prediction : Tensor
The prediction.
target : Tensor
The target. The shape should be broadcastable to ``prediction``.
mask : Tensor
The mask. The shape should be broadcatable to the broadcasted shape of
``prediction`` and ``target``.
Returns
-------
Tensor [shape=(1,)]
The masked L1 loss.
"""
abs_error = F.l1_loss(prediction, target, reduction='none')
loss = weighted_mean(abs_error, mask)
return loss
def masked_softmax_with_cross_entropy(logits, label, mask, axis=-1):
"""Compute masked softmax with cross entropy loss.
Parameters
----------
logits : Tensor
The logits. The ``axis``-th axis is the class dimension.
label : Tensor [dtype: int]
The label. The size of the ``axis``-th axis should be 1.
mask : Tensor
The mask. The shape should be broadcastable to ``label``.
axis : int, optional
The index of the class dimension in the shape of ``logits``, by default
-1.
Returns
-------
Tensor [shape=(1,)]
The masked softmax with cross entropy loss.
"""
ce = F.softmax_with_cross_entropy(logits, label, axis=axis)
loss = weighted_mean(ce, mask)
return loss