wehub-resource-sync
100 lines
3.3 KiB
C++
100 lines
3.3 KiB
C++
// Copyright (c) 2023 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.
|
|
|
|
#pragma once
|
|
|
|
#include <vector>
|
|
#include "paddle/phi/common/type_traits.h"
|
|
#include "paddle/phi/core/dense_tensor.h"
|
|
#include "paddle/phi/kernels/cpu/elementwise.h"
|
|
#include "paddle/phi/kernels/funcs/elementwise_base.h"
|
|
#include "paddle/phi/kernels/funcs/elementwise_functor.h"
|
|
#include "paddle/phi/kernels/funcs/fft.h"
|
|
#include "paddle/phi/kernels/funcs/fft_fill_conj.h"
|
|
#include "paddle/phi/kernels/funcs/frame_functor.h"
|
|
|
|
namespace phi {
|
|
|
|
template <typename T, typename Context>
|
|
void StftKernel(const Context& dev_ctx,
|
|
const DenseTensor& x,
|
|
const DenseTensor& window,
|
|
int n_fft,
|
|
int hop_length,
|
|
bool normalized,
|
|
bool onesided,
|
|
DenseTensor* out) {
|
|
using C = dtype::complex<T>;
|
|
|
|
dev_ctx.template Alloc<C>(out);
|
|
|
|
const size_t x_rank = x.dims().size();
|
|
const size_t out_rank = out->dims().size();
|
|
|
|
const size_t n_frames = out->dims()[out_rank - 1];
|
|
const size_t seq_length = x.dims()[x_rank - 1];
|
|
|
|
std::vector<int64_t> axes = {1};
|
|
|
|
// Frame
|
|
DenseTensor frames;
|
|
DDim frames_dims(out->dims());
|
|
frames_dims.at(axes.back()) = n_fft;
|
|
frames.Resize(frames_dims);
|
|
dev_ctx.template Alloc<T>(&frames);
|
|
funcs::FrameFunctor<Context, T>()(dev_ctx,
|
|
&x,
|
|
&frames,
|
|
seq_length,
|
|
n_fft,
|
|
n_frames,
|
|
hop_length,
|
|
/*is_grad*/ false);
|
|
|
|
// Window
|
|
DenseTensor frames_w;
|
|
frames_w.Resize(frames_dims);
|
|
dev_ctx.template Alloc<T>(&frames_w);
|
|
funcs::ElementwiseCompute<funcs::MultiplyFunctor<T>, T, T>(
|
|
dev_ctx,
|
|
frames,
|
|
window,
|
|
funcs::MultiplyFunctor<T>(),
|
|
&frames_w,
|
|
axes.back());
|
|
|
|
// FFTR2C
|
|
funcs::FFTNormMode normalization;
|
|
if (normalized) {
|
|
normalization = funcs::get_norm_from_string("ortho", true);
|
|
} else {
|
|
normalization = funcs::get_norm_from_string("backward", true);
|
|
}
|
|
funcs::FFTR2CFunctor<Context, T, C> fft_r2c_func;
|
|
|
|
if (onesided) {
|
|
fft_r2c_func(dev_ctx, frames_w, out, axes, normalization, true);
|
|
} else {
|
|
DDim onesided_dims(out->dims());
|
|
const int64_t onesided_axis_size = out->dims().at(axes.back()) / 2 + 1;
|
|
onesided_dims.at(axes.back()) = onesided_axis_size;
|
|
DenseTensor onesided_out;
|
|
onesided_out.Resize(onesided_dims);
|
|
dev_ctx.template Alloc<T>(&onesided_out);
|
|
fft_r2c_func(dev_ctx, frames_w, &onesided_out, axes, normalization, true);
|
|
funcs::FFTFillConj<Context, C>(dev_ctx, &onesided_out, out, axes);
|
|
}
|
|
}
|
|
} // namespace phi
|