add multi-band melgan finetune scripts

pull/995/head
TianYuan 3 years ago
parent dc7daa2a61
commit a6ac497f8e

@ -13,7 +13,6 @@
# limitations under the License.
import argparse
from pathlib import Path
from typing import Union
import numpy as np
import paddle
@ -23,129 +22,12 @@ from yacs.config import CfgNode
from paddlespeech.t2s.frontend.zh_frontend import Frontend
from paddlespeech.t2s.models.fastspeech2 import FastSpeech2
from paddlespeech.t2s.models.fastspeech2 import FastSpeech2Inference
from paddlespeech.t2s.models.fastspeech2 import StyleFastSpeech2Inference
from paddlespeech.t2s.models.parallel_wavegan import PWGGenerator
from paddlespeech.t2s.models.parallel_wavegan import PWGInference
from paddlespeech.t2s.modules.normalizer import ZScore
class StyleFastSpeech2Inference(FastSpeech2Inference):
def __init__(self, normalizer, model, pitch_stats_path, energy_stats_path):
super().__init__(normalizer, model)
pitch_mean, pitch_std = np.load(pitch_stats_path)
self.pitch_mean = paddle.to_tensor(pitch_mean)
self.pitch_std = paddle.to_tensor(pitch_std)
energy_mean, energy_std = np.load(energy_stats_path)
self.energy_mean = paddle.to_tensor(energy_mean)
self.energy_std = paddle.to_tensor(energy_std)
def denorm(self, data, mean, std):
return data * std + mean
def norm(self, data, mean, std):
return (data - mean) / std
def forward(self,
text: paddle.Tensor,
durations: Union[paddle.Tensor, np.ndarray]=None,
durations_scale: Union[int, float]=None,
durations_bias: Union[int, float]=None,
pitch: Union[paddle.Tensor, np.ndarray]=None,
pitch_scale: Union[int, float]=None,
pitch_bias: Union[int, float]=None,
energy: Union[paddle.Tensor, np.ndarray]=None,
energy_scale: Union[int, float]=None,
energy_bias: Union[int, float]=None,
robot: bool=False):
"""
Parameters
----------
text : Tensor(int64)
Input sequence of characters (T,).
speech : Tensor, optional
Feature sequence to extract style (N, idim).
durations : paddle.Tensor/np.ndarray, optional (int64)
Groundtruth of duration (T,), this will overwrite the set of durations_scale and durations_bias
durations_scale: int/float, optional
durations_bias: int/float, optional
pitch : paddle.Tensor/np.ndarray, optional
Groundtruth of token-averaged pitch (T, 1), this will overwrite the set of pitch_scale and pitch_bias
pitch_scale: int/float, optional
In denormed HZ domain.
pitch_bias: int/float, optional
In denormed HZ domain.
energy : paddle.Tensor/np.ndarray, optional
Groundtruth of token-averaged energy (T, 1), this will overwrite the set of energy_scale and energy_bias
energy_scale: int/float, optional
In denormed domain.
energy_bias: int/float, optional
In denormed domain.
robot : bool, optional
Weather output robot style
Returns
----------
Tensor
Output sequence of features (L, odim).
"""
normalized_mel, d_outs, p_outs, e_outs = self.acoustic_model.inference(
text, durations=None, pitch=None, energy=None)
# priority: groundtruth > scale/bias > previous output
# set durations
if isinstance(durations, np.ndarray):
durations = paddle.to_tensor(durations)
elif isinstance(durations, paddle.Tensor):
durations = durations
elif durations_scale or durations_bias:
durations_scale = durations_scale if durations_scale is not None else 1
durations_bias = durations_bias if durations_bias is not None else 0
durations = durations_scale * d_outs + durations_bias
else:
durations = d_outs
if robot:
# set normed pitch to zeros have the same effect with set denormd ones to mean
pitch = paddle.zeros(p_outs.shape)
# set pitch, can overwrite robot set
if isinstance(pitch, np.ndarray):
pitch = paddle.to_tensor(pitch)
elif isinstance(pitch, paddle.Tensor):
pitch = pitch
elif pitch_scale or pitch_bias:
pitch_scale = pitch_scale if pitch_scale is not None else 1
pitch_bias = pitch_bias if pitch_bias is not None else 0
p_Hz = paddle.exp(
self.denorm(p_outs, self.pitch_mean, self.pitch_std))
p_HZ = pitch_scale * p_Hz + pitch_bias
pitch = self.norm(paddle.log(p_HZ), self.pitch_mean, self.pitch_std)
else:
pitch = p_outs
# set energy
if isinstance(energy, np.ndarray):
energy = paddle.to_tensor(energy)
elif isinstance(energy, paddle.Tensor):
energy = energy
elif energy_scale or energy_bias:
energy_scale = energy_scale if energy_scale is not None else 1
energy_bias = energy_bias if energy_bias is not None else 0
e_dnorm = self.denorm(e_outs, self.energy_mean, self.energy_std)
e_dnorm = energy_scale * e_dnorm + energy_bias
energy = self.norm(e_dnorm, self.energy_mean, self.energy_std)
else:
energy = e_outs
normalized_mel, d_outs, p_outs, e_outs = self.acoustic_model.inference(
text,
durations=durations,
pitch=pitch,
energy=energy,
use_teacher_forcing=True)
logmel = self.normalizer.inverse(normalized_mel)
return logmel
def evaluate(args, fastspeech2_config, pwg_config):
# construct dataset for evaluation

@ -0,0 +1,139 @@
# This is the hyperparameter configuration file for MelGAN.
# Please make sure this is adjusted for the CSMSC dataset. If you want to
# apply to the other dataset, you might need to carefully change some parameters.
# This configuration requires ~ 8GB memory and will finish within 7 days on Titan V.
# This configuration is based on full-band MelGAN but the hop size and sampling
# rate is different from the paper (16kHz vs 24kHz). The number of iteraions
# is not shown in the paper so currently we train 1M iterations (not sure enough
# to converge). The optimizer setting is based on @dathudeptrai advice.
# https://github.com/kan-bayashi/ParallelWaveGAN/issues/143#issuecomment-632539906
###########################################################
# FEATURE EXTRACTION SETTING #
###########################################################
fs: 24000 # Sampling rate.
n_fft: 2048 # FFT size. (in samples)
n_shift: 300 # Hop size. (in samples)
win_length: 1200 # Window length. (in samples)
# If set to null, it will be the same as fft_size.
window: "hann" # Window function.
n_mels: 80 # Number of mel basis.
fmin: 80 # Minimum freq in mel basis calculation. (Hz)
fmax: 7600 # Maximum frequency in mel basis calculation. (Hz)
###########################################################
# GENERATOR NETWORK ARCHITECTURE SETTING #
###########################################################
generator_params:
in_channels: 80 # Number of input channels.
out_channels: 4 # Number of output channels.
kernel_size: 7 # Kernel size of initial and final conv layers.
channels: 384 # Initial number of channels for conv layers.
upsample_scales: [5, 5, 3] # List of Upsampling scales.
stack_kernel_size: 3 # Kernel size of dilated conv layers in residual stack.
stacks: 4 # Number of stacks in a single residual stack module.
use_weight_norm: True # Whether to use weight normalization.
use_causal_conv: False # Whether to use causal convolution.
use_final_nonlinear_activation: True
###########################################################
# DISCRIMINATOR NETWORK ARCHITECTURE SETTING #
###########################################################
discriminator_params:
in_channels: 1 # Number of input channels.
out_channels: 1 # Number of output channels.
scales: 3 # Number of multi-scales.
downsample_pooling: "AvgPool1D" # Pooling type for the input downsampling.
downsample_pooling_params: # Parameters of the above pooling function.
kernel_size: 4
stride: 2
padding: 1
exclusive: True
kernel_sizes: [5, 3] # List of kernel size.
channels: 16 # Number of channels of the initial conv layer.
max_downsample_channels: 512 # Maximum number of channels of downsampling layers.
downsample_scales: [4, 4, 4] # List of downsampling scales.
nonlinear_activation: "LeakyReLU" # Nonlinear activation function.
nonlinear_activation_params: # Parameters of nonlinear activation function.
negative_slope: 0.2
use_weight_norm: True # Whether to use weight norm.
###########################################################
# STFT LOSS SETTING #
###########################################################
use_stft_loss: true
stft_loss_params:
fft_sizes: [1024, 2048, 512] # List of FFT size for STFT-based loss.
hop_sizes: [120, 240, 50] # List of hop size for STFT-based loss
win_lengths: [600, 1200, 240] # List of window length for STFT-based loss.
window: "hann" # Window function for STFT-based loss
use_subband_stft_loss: true
subband_stft_loss_params:
fft_sizes: [384, 683, 171] # List of FFT size for STFT-based loss.
hop_sizes: [30, 60, 10] # List of hop size for STFT-based loss
win_lengths: [150, 300, 60] # List of window length for STFT-based loss.
window: "hann" # Window function for STFT-based loss
###########################################################
# ADVERSARIAL LOSS SETTING #
###########################################################
use_feat_match_loss: false # Whether to use feature matching loss.
lambda_adv: 2.5 # Loss balancing coefficient for adversarial loss.
###########################################################
# DATA LOADER SETTING #
###########################################################
batch_size: 64 # Batch size.
batch_max_steps: 16200 # Length of each audio in batch. Make sure dividable by hop_size.
num_workers: 2 # Number of workers in DataLoader.
###########################################################
# OPTIMIZER & SCHEDULER SETTING #
###########################################################
generator_optimizer_params:
epsilon: 1.0e-7 # Generator's epsilon.
weight_decay: 0.0 # Generator's weight decay coefficient.
generator_grad_norm: -1 # Generator's gradient norm.
generator_scheduler_params:
learning_rate: 1.0e-3 # Generator's learning rate.
gamma: 0.5 # Generator's scheduler gamma.
milestones: # At each milestone, lr will be multiplied by gamma.
- 100000
- 200000
- 300000
- 400000
- 500000
- 600000
discriminator_optimizer_params:
epsilon: 1.0e-7 # Discriminator's epsilon.
weight_decay: 0.0 # Discriminator's weight decay coefficient.
discriminator_grad_norm: -1 # Discriminator's gradient norm.
discriminator_scheduler_params:
learning_rate: 1.0e-3 # Discriminator's learning rate.
gamma: 0.5 # Discriminator's scheduler gamma.
milestones: # At each milestone, lr will be multiplied by gamma.
- 100000
- 200000
- 300000
- 400000
- 500000
- 600000
###########################################################
# INTERVAL SETTING #
###########################################################
discriminator_train_start_steps: 200000 # Number of steps to start to train discriminator.
train_max_steps: 1200000 # Number of training steps.
save_interval_steps: 1000 # Interval steps to save checkpoint.
eval_interval_steps: 1000 # Interval steps to evaluate the network.
###########################################################
# OTHER SETTING #
###########################################################
num_snapshots: 10 # max number of snapshots to keep while training
seed: 42 # random seed for paddle, random, and np.random

@ -0,0 +1,65 @@
#!/bin/bash
source path.sh
gpus=0
stage=0
stop_stage=100
source ${MAIN_ROOT}/utils/parse_options.sh || exit 1
if [ ${stage} -le 0 ] && [ ${stop_stage} -ge 0 ]; then
python3 ${MAIN_ROOT}/paddlespeech/t2s/exps/fastspeech2/gen_gt_duration_mel.py \
--fastspeech2-config=fastspeech2_nosil_baker_ckpt_0.4/default.yaml \
--fastspeech2-checkpoint=fastspeech2_nosil_baker_ckpt_0.4/snapshot_iter_76000.pdz \
--fastspeech2-stat=fastspeech2_nosil_baker_ckpt_0.4/speech_stats.npy \
--dur-file=durations.txt \
--output-dir=dump_finetune \
--phones-dict=fastspeech2_nosil_baker_ckpt_0.4/phone_id_map.txt
fi
if [ ${stage} -le 1 ] && [ ${stop_stage} -ge 1 ]; then
python3 local/link_wav.py \
--old-dump-dir=dump \
--dump-dir=dump_finetune
fi
if [ ${stage} -le 2 ] && [ ${stop_stage} -ge 2 ]; then
# get features' stats(mean and std)
echo "Get features' stats ..."
python3 ${MAIN_ROOT}/utils/compute_statistics.py \
--metadata=dump_finetune/train/raw/metadata.jsonl \
--field-name="feats"
fi
if [ ${stage} -le 3 ] && [ ${stop_stage} -ge 3 ]; then
# normalize, dev and test should use train's stats
echo "Normalize ..."
python3 ${BIN_DIR}/../normalize.py \
--metadata=dump_finetune/train/raw/metadata.jsonl \
--dumpdir=dump_finetune/train/norm \
--stats=dump_finetune/train/feats_stats.npy
python3 ${BIN_DIR}/../normalize.py \
--metadata=dump_finetune/dev/raw/metadata.jsonl \
--dumpdir=dump_finetune/dev/norm \
--stats=dump_finetune/train/feats_stats.npy
python3 ${BIN_DIR}/../normalize.py \
--metadata=dump_finetune/test/raw/metadata.jsonl \
--dumpdir=dump_finetune/test/norm \
--stats=dump_finetune/train/feats_stats.npy
fi
if [ ${stage} -le 4 ] && [ ${stop_stage} -ge 4 ]; then
CUDA_VISIBLE_DEVICES=${gpus} \
FLAGS_cudnn_exhaustive_search=true \
FLAGS_conv_workspace_size_limit=4000 \
python ${BIN_DIR}/train.py \
--train-metadata=dump_finetune/train/norm/metadata.jsonl \
--dev-metadata=dump_finetune/dev/norm/metadata.jsonl \
--config=conf/finetune.yaml \
--output-dir=exp/finetune \
--ngpu=1
fi

@ -0,0 +1,85 @@
# Copyright (c) 2021 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 argparse
import os
from operator import itemgetter
from pathlib import Path
import jsonlines
import numpy as np
def main():
# parse config and args
parser = argparse.ArgumentParser(
description="Preprocess audio and then extract features .")
parser.add_argument(
"--old-dump-dir",
default=None,
type=str,
help="directory to dump feature files.")
parser.add_argument(
"--dump-dir",
type=str,
required=True,
help="directory to finetune dump feature files.")
args = parser.parse_args()
old_dump_dir = Path(args.old_dump_dir).expanduser()
old_dump_dir = old_dump_dir.resolve()
dump_dir = Path(args.dump_dir).expanduser()
# use absolute path
dump_dir = dump_dir.resolve()
dump_dir.mkdir(parents=True, exist_ok=True)
assert old_dump_dir.is_dir()
assert dump_dir.is_dir()
for sub in ["train", "dev", "test"]:
# 把 old_dump_dir 里面的 *-wave.npy 软连接到 dump_dir 的对应位置
output_dir = dump_dir / sub
output_dir.mkdir(parents=True, exist_ok=True)
results = []
for name in os.listdir(output_dir / "raw"):
# 003918_feats.npy
utt_id = name.split("_")[0]
mel_path = output_dir / ("raw/" + name)
gen_mel = np.load(mel_path)
wave_name = utt_id + "_wave.npy"
wav = np.load(old_dump_dir / sub / ("raw/" + wave_name))
os.symlink(old_dump_dir / sub / ("raw/" + wave_name),
output_dir / ("raw/" + wave_name))
num_sample = wav.shape[0]
num_frames = gen_mel.shape[0]
wav_path = output_dir / ("raw/" + wave_name)
record = {
"utt_id": utt_id,
"num_samples": num_sample,
"num_frames": num_frames,
"feats": str(mel_path),
"wave": str(wav_path),
}
results.append(record)
results.sort(key=itemgetter("utt_id"))
with jsonlines.open(output_dir / "raw/metadata.jsonl", 'w') as writer:
for item in results:
writer.write(item)
if __name__ == "__main__":
main()

@ -110,10 +110,10 @@ class Clip(object):
if len(x) < c.shape[0] * self.hop_size:
x = np.pad(x, (0, c.shape[0] * self.hop_size - len(x)), mode="edge")
elif len(x) > c.shape[0] * self.hop_size:
print(
f"wave length: ({len(x)}), mel length: ({c.shape[0]}), hop size: ({self.hop_size })"
)
x = x[:c.shape[1] * self.hop_size]
# print(
# f"wave length: ({len(x)}), mel length: ({c.shape[0]}), hop size: ({self.hop_size })"
# )
x = x[:c.shape[0] * self.hop_size]
# check the legnth is valid
assert len(x) == c.shape[

@ -0,0 +1,167 @@
# Copyright (c) 2021 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.
# generate mels using durations.txt
# for mb melgan finetune
# 长度和原本的 mel 不一致怎么办?
import argparse
from pathlib import Path
import numpy as np
import paddle
import yaml
from yacs.config import CfgNode
from paddlespeech.t2s.datasets.preprocess_utils import get_phn_dur
from paddlespeech.t2s.datasets.preprocess_utils import merge_silence
from paddlespeech.t2s.models.fastspeech2 import FastSpeech2
from paddlespeech.t2s.models.fastspeech2 import StyleFastSpeech2Inference
from paddlespeech.t2s.modules.normalizer import ZScore
def evaluate(args, fastspeech2_config):
# construct dataset for evaluation
with open(args.phones_dict, "r") as f:
phn_id = [line.strip().split() for line in f.readlines()]
vocab_size = len(phn_id)
print("vocab_size:", vocab_size)
phone_dict = {}
for phn, id in phn_id:
phone_dict[phn] = int(id)
odim = fastspeech2_config.n_mels
model = FastSpeech2(
idim=vocab_size, odim=odim, **fastspeech2_config["model"])
model.set_state_dict(
paddle.load(args.fastspeech2_checkpoint)["main_params"])
model.eval()
stat = np.load(args.fastspeech2_stat)
mu, std = stat
mu = paddle.to_tensor(mu)
std = paddle.to_tensor(std)
fastspeech2_normalizer = ZScore(mu, std)
fastspeech2_inference = StyleFastSpeech2Inference(fastspeech2_normalizer,
model)
fastspeech2_inference.eval()
output_dir = Path(args.output_dir)
output_dir.mkdir(parents=True, exist_ok=True)
sentences, speaker_set = get_phn_dur(args.dur_file)
merge_silence(sentences)
for i, utt_id in enumerate(sentences):
phones = sentences[utt_id][0]
durations = sentences[utt_id][1]
speaker = sentences[utt_id][2]
# 裁剪掉开头和结尾的 sil
if args.cut_sil:
if phones[0] == "sil" and len(durations) > 1:
durations = durations[1:]
phones = phones[1:]
if phones[-1] == 'sil' and len(durations) > 1:
durations = durations[:-1]
phones = phones[:-1]
# sentences[utt_id][0] = phones
# sentences[utt_id][1] = durations
phone_ids = [phone_dict[phn] for phn in phones]
phone_ids = paddle.to_tensor(np.array(phone_ids))
durations = paddle.to_tensor(np.array(durations))
# 生成的和真实的可能有 1, 2 帧的差距,但是 batch_fn 会修复
# split data into 3 sections
if args.dataset == "baker":
num_train = 9800
num_dev = 100
if i in range(0, num_train):
sub_output_dir = output_dir / ("train/raw")
elif i in range(num_train, num_train + num_dev):
sub_output_dir = output_dir / ("dev/raw")
else:
sub_output_dir = output_dir / ("test/raw")
sub_output_dir.mkdir(parents=True, exist_ok=True)
with paddle.no_grad():
mel = fastspeech2_inference(phone_ids, durations=durations)
np.save(sub_output_dir / (utt_id + "_feats.npy"), mel)
def main():
# parse args and config and redirect to train_sp
parser = argparse.ArgumentParser(
description="Synthesize with fastspeech2 & parallel wavegan.")
parser.add_argument(
"--dataset",
default="baker",
type=str,
help="name of dataset, should in {baker, ljspeech, vctk} now")
parser.add_argument(
"--fastspeech2-config", type=str, help="fastspeech2 config file.")
parser.add_argument(
"--fastspeech2-checkpoint",
type=str,
help="fastspeech2 checkpoint to load.")
parser.add_argument(
"--fastspeech2-stat",
type=str,
help="mean and standard deviation used to normalize spectrogram when training fastspeech2."
)
parser.add_argument(
"--phones-dict",
type=str,
default="phone_id_map.txt",
help="phone vocabulary file.")
parser.add_argument(
"--dur-file", default=None, type=str, help="path to durations.txt.")
parser.add_argument("--output-dir", type=str, help="output dir.")
parser.add_argument(
"--ngpu", type=int, default=1, help="if ngpu == 0, use cpu.")
parser.add_argument("--verbose", type=int, default=1, help="verbose.")
def str2bool(str):
return True if str.lower() == 'true' else False
parser.add_argument(
"--cut-sil",
type=str2bool,
default=True,
help="whether cut sil in the edge of audio")
args = parser.parse_args()
if args.ngpu == 0:
paddle.set_device("cpu")
elif args.ngpu > 0:
paddle.set_device("gpu")
else:
print("ngpu should >= 0 !")
with open(args.fastspeech2_config) as f:
fastspeech2_config = CfgNode(yaml.safe_load(f))
print("========Args========")
print(yaml.safe_dump(vars(args)))
print("========Config========")
print(fastspeech2_config)
evaluate(args, fastspeech2_config)
if __name__ == "__main__":
main()

@ -16,7 +16,9 @@
from typing import Dict
from typing import Sequence
from typing import Tuple
from typing import Union
import numpy as np
import paddle
import paddle.nn.functional as F
from paddle import nn
@ -687,6 +689,129 @@ class FastSpeech2Inference(nn.Layer):
return logmel
class StyleFastSpeech2Inference(FastSpeech2Inference):
def __init__(self,
normalizer,
model,
pitch_stats_path=None,
energy_stats_path=None):
super().__init__(normalizer, model)
if pitch_stats_path:
pitch_mean, pitch_std = np.load(pitch_stats_path)
self.pitch_mean = paddle.to_tensor(pitch_mean)
self.pitch_std = paddle.to_tensor(pitch_std)
if energy_stats_path:
energy_mean, energy_std = np.load(energy_stats_path)
self.energy_mean = paddle.to_tensor(energy_mean)
self.energy_std = paddle.to_tensor(energy_std)
def denorm(self, data, mean, std):
return data * std + mean
def norm(self, data, mean, std):
return (data - mean) / std
def forward(self,
text: paddle.Tensor,
durations: Union[paddle.Tensor, np.ndarray]=None,
durations_scale: Union[int, float]=None,
durations_bias: Union[int, float]=None,
pitch: Union[paddle.Tensor, np.ndarray]=None,
pitch_scale: Union[int, float]=None,
pitch_bias: Union[int, float]=None,
energy: Union[paddle.Tensor, np.ndarray]=None,
energy_scale: Union[int, float]=None,
energy_bias: Union[int, float]=None,
robot: bool=False):
"""
Parameters
----------
text : Tensor(int64)
Input sequence of characters (T,).
speech : Tensor, optional
Feature sequence to extract style (N, idim).
durations : paddle.Tensor/np.ndarray, optional (int64)
Groundtruth of duration (T,), this will overwrite the set of durations_scale and durations_bias
durations_scale: int/float, optional
durations_bias: int/float, optional
pitch : paddle.Tensor/np.ndarray, optional
Groundtruth of token-averaged pitch (T, 1), this will overwrite the set of pitch_scale and pitch_bias
pitch_scale: int/float, optional
In denormed HZ domain.
pitch_bias: int/float, optional
In denormed HZ domain.
energy : paddle.Tensor/np.ndarray, optional
Groundtruth of token-averaged energy (T, 1), this will overwrite the set of energy_scale and energy_bias
energy_scale: int/float, optional
In denormed domain.
energy_bias: int/float, optional
In denormed domain.
robot : bool, optional
Weather output robot style
Returns
----------
Tensor
Output sequence of features (L, odim).
"""
normalized_mel, d_outs, p_outs, e_outs = self.acoustic_model.inference(
text, durations=None, pitch=None, energy=None)
# priority: groundtruth > scale/bias > previous output
# set durations
if isinstance(durations, np.ndarray):
durations = paddle.to_tensor(durations)
elif isinstance(durations, paddle.Tensor):
durations = durations
elif durations_scale or durations_bias:
durations_scale = durations_scale if durations_scale is not None else 1
durations_bias = durations_bias if durations_bias is not None else 0
durations = durations_scale * d_outs + durations_bias
else:
durations = d_outs
if robot:
# set normed pitch to zeros have the same effect with set denormd ones to mean
pitch = paddle.zeros(p_outs.shape)
# set pitch, can overwrite robot set
if isinstance(pitch, np.ndarray):
pitch = paddle.to_tensor(pitch)
elif isinstance(pitch, paddle.Tensor):
pitch = pitch
elif pitch_scale or pitch_bias:
pitch_scale = pitch_scale if pitch_scale is not None else 1
pitch_bias = pitch_bias if pitch_bias is not None else 0
p_Hz = paddle.exp(
self.denorm(p_outs, self.pitch_mean, self.pitch_std))
p_HZ = pitch_scale * p_Hz + pitch_bias
pitch = self.norm(paddle.log(p_HZ), self.pitch_mean, self.pitch_std)
else:
pitch = p_outs
# set energy
if isinstance(energy, np.ndarray):
energy = paddle.to_tensor(energy)
elif isinstance(energy, paddle.Tensor):
energy = energy
elif energy_scale or energy_bias:
energy_scale = energy_scale if energy_scale is not None else 1
energy_bias = energy_bias if energy_bias is not None else 0
e_dnorm = self.denorm(e_outs, self.energy_mean, self.energy_std)
e_dnorm = energy_scale * e_dnorm + energy_bias
energy = self.norm(e_dnorm, self.energy_mean, self.energy_std)
else:
energy = e_outs
normalized_mel, d_outs, p_outs, e_outs = self.acoustic_model.inference(
text,
durations=durations,
pitch=pitch,
energy=energy,
use_teacher_forcing=True)
logmel = self.normalizer.inverse(normalized_mel)
return logmel
class FastSpeech2Loss(nn.Layer):
"""Loss function module for FastSpeech2."""

Loading…
Cancel
Save