• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
1 /**
2  * Copyright 2021 Huawei Technologies Co., Ltd
3  *
4  * Licensed under the Apache License, Version 2.0 (the "License");
5  * you may not use this file except in compliance with the License.
6  * You may obtain a copy of the License at
7  *
8  * http://www.apache.org/licenses/LICENSE-2.0
9  *
10  * Unless required by applicable law or agreed to in writing, software
11  * distributed under the License is distributed on an "AS IS" BASIS,
12  * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
13  * See the License for the specific language governing permissions and
14  * limitations under the License.
15  */
16 
17 #include "nnacl/infer/audio_spectrogram_infer.h"
18 #include "nnacl/infer/infer_register.h"
19 
Log2Ceil(unsigned length)20 unsigned Log2Ceil(unsigned length) {
21   if (length == 0) {
22     return 0;
23   }
24   int floor = 0;
25   for (int i = 4; i >= 0; --i) {
26     const unsigned shift = (1 << (unsigned)i);
27     unsigned tmp = length >> shift;
28     if (tmp != 0) {
29       length = tmp;
30       floor += shift;
31     }
32   }
33   return length == (length & ~(length - 1)) ? floor : floor + 1;
34 }
35 
GetFftLength(unsigned length)36 unsigned GetFftLength(unsigned length) {
37   unsigned shift = Log2Ceil(length);
38   return 1 << shift;
39 }
40 
AudioSpectrogramInferShape(const TensorC * const * inputs,size_t inputs_size,TensorC ** outputs,size_t outputs_size,OpParameter * parameter)41 int AudioSpectrogramInferShape(const TensorC *const *inputs, size_t inputs_size, TensorC **outputs, size_t outputs_size,
42                                OpParameter *parameter) {
43   int check_ret = CheckAugmentWithMinSize(inputs, inputs_size, outputs, outputs_size, parameter, 1, 1);
44   if (check_ret != NNACL_OK) {
45     return check_ret;
46   }
47 
48   const TensorC *input = inputs[0];
49   TensorC *output = outputs[0];
50   SetDataTypeFormat(output, input);
51   if (!InferFlag(inputs, inputs_size)) {
52     return NNACL_INFER_INVALID;
53   }
54   if (input->shape_size_ != 2) {
55     return NNACL_ERR;
56   }
57   AudioSpectrogramParameter *param = (AudioSpectrogramParameter *)parameter;
58   if (param->window_size_ < 2) {
59     return NNACL_ERR;
60   }
61   if (param->stride_ < 1) {
62     return NNACL_ERR;
63   }
64   int output_shape[3];
65   output_shape[0] = input->shape_[1];
66   int sample_sub_window = input->shape_[0] - param->window_size_;
67   output_shape[1] = sample_sub_window < 0 ? 0 : 1 + sample_sub_window / param->stride_;
68   // compute fft length
69   int fft_length = (int)GetFftLength(param->window_size_);
70   output_shape[2] = fft_length / 2 + 1;
71   SetShapeArray(output, output_shape, 3);
72   return NNACL_OK;
73 }
74 
75 REG_INFER(AudioSpectrogram, PrimType_AudioSpectrogram, AudioSpectrogramInferShape)
76