1/* Copyright 2017 The TensorFlow Authors. All Rights Reserved.
2
3Licensed under the Apache License, Version 2.0 (the "License");
4you may not use this file except in compliance with the License.
5You may obtain a copy of the License at
6
7 http://www.apache.org/licenses/LICENSE-2.0
8
9Unless required by applicable law or agreed to in writing, software
10distributed under the License is distributed on an "AS IS" BASIS,
11WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
12See the License for the specific language governing permissions and
13limitations under the License.
14==============================================================================*/
15
16// See docs in ../ops/audio_ops.cc
17
18#include "tensorflow/core/framework/op_kernel.h"
19#include "tensorflow/core/framework/register_types.h"
20#include "tensorflow/core/framework/tensor.h"
21#include "tensorflow/core/framework/tensor_shape.h"
22#include "tensorflow/core/framework/types.h"
23#include "tensorflow/core/lib/core/status.h"
24#include "tensorflow/core/lib/wav/wav_io.h"
25
26namespace tensorflow {
27
28// Decode the contents of a WAV file
29class DecodeWavOp : public OpKernel {
30 public:
31 explicit DecodeWavOp(OpKernelConstruction* context) : OpKernel(context) {
32 OP_REQUIRES_OK(context,
33 context->GetAttr("desired_channels", &desired_channels_));
34 OP_REQUIRES_OK(context,
35 context->GetAttr("desired_samples", &desired_samples_));
36 }
37
38 void Compute(OpKernelContext* context) override {
39 const Tensor& contents = context->input(0);
40 OP_REQUIRES(context, TensorShapeUtils::IsScalar(contents.shape()),
41 errors::InvalidArgument("contents must be scalar, got shape ",
42 contents.shape().DebugString()));
43 const string& wav_string = contents.scalar<tstring>()();
44 OP_REQUIRES(context, wav_string.size() <= std::numeric_limits<int>::max(),
45 errors::InvalidArgument("WAV contents are too large for int: ",
46 wav_string.size()));
47
48 std::vector<float> decoded_samples;
49 uint32 decoded_sample_count;
50 uint16 decoded_channel_count;
51 uint32 decoded_sample_rate;
52 OP_REQUIRES_OK(context,
53 wav::DecodeLin16WaveAsFloatVector(
54 wav_string, &decoded_samples, &decoded_sample_count,
55 &decoded_channel_count, &decoded_sample_rate));
56
57 int32_t output_sample_count;
58 if (desired_samples_ == -1) {
59 output_sample_count = decoded_sample_count;
60 } else {
61 output_sample_count = desired_samples_;
62 }
63 int32_t output_channel_count;
64 if (desired_channels_ == -1) {
65 output_channel_count = decoded_channel_count;
66 } else {
67 output_channel_count = desired_channels_;
68 }
69
70 Tensor* output = nullptr;
71 OP_REQUIRES_OK(
72 context,
73 context->allocate_output(
74 0, TensorShape({output_sample_count, output_channel_count}),
75 &output));
76
77 auto output_matrix = output->matrix<float>();
78 for (int sample = 0; sample < output_sample_count; ++sample) {
79 for (int channel = 0; channel < output_channel_count; ++channel) {
80 float output_value;
81 if (sample >= decoded_sample_count) {
82 output_value = 0.0f;
83 } else {
84 int source_channel;
85 if (channel < decoded_channel_count) {
86 source_channel = channel;
87 } else {
88 source_channel = decoded_channel_count - 1;
89 }
90 const int decoded_index =
91 (sample * decoded_channel_count) + source_channel;
92 output_value = decoded_samples[decoded_index];
93 }
94 output_matrix(sample, channel) = output_value;
95 }
96 }
97
98 Tensor* sample_rate_output = nullptr;
99 OP_REQUIRES_OK(context, context->allocate_output(1, TensorShape({}),
100 &sample_rate_output));
101 sample_rate_output->flat<int32>()(0) = decoded_sample_rate;
102 }
103
104 private:
105 int32 desired_channels_;
106 int32 desired_samples_;
107};
108REGISTER_KERNEL_BUILDER(Name("DecodeWav").Device(DEVICE_CPU), DecodeWavOp);
109
110} // namespace tensorflow
111