1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
|
#include "resample.hpp"
#include <stdint.h>
#include <stdlib.h>
#include <string.h>
#include <algorithm>
#include <cmath>
#include <numeric>
#include "esp_log.h"
#include "sample.hpp"
#include "stream_info.hpp"
namespace audio {
static constexpr char kTag[] = "resample";
static constexpr double kLowPassRatio = 0.5;
static constexpr size_t kNumFilters = 8;
static constexpr size_t kTapsPerFilter = 8;
typedef std::array<float, kTapsPerFilter> Filter;
static std::array<Filter, kNumFilters + 1> sFilters{};
static bool sFiltersInitialised = false;
auto InitFilter(int index) -> void;
Resampler::Resampler(uint32_t source_sample_rate,
uint32_t target_sample_rate,
uint8_t num_channels)
: source_sample_rate_(source_sample_rate),
target_sample_rate_(target_sample_rate),
factor_(static_cast<double>(target_sample_rate) /
static_cast<double>(source_sample_rate)),
num_channels_(num_channels) {
channel_buffers_.resize(num_channels);
channel_buffer_size_ = kTapsPerFilter * 16;
for (int i = 0; i < num_channels; i++) {
channel_buffers_[i] =
static_cast<float*>(calloc(sizeof(float), channel_buffer_size_));
}
output_offset_ = kTapsPerFilter / 2.0f;
input_index_ = kTapsPerFilter;
if (!sFiltersInitialised) {
sFiltersInitialised = true;
for (int i = 0; i < kNumFilters + 1; i++) {
InitFilter(i);
}
}
}
Resampler::~Resampler() {}
auto Resampler::Process(cpp::span<const sample::Sample> input,
cpp::span<sample::Sample> output,
bool end_of_data) -> std::pair<size_t, size_t> {
size_t samples_used = 0;
size_t samples_produced = 0;
size_t input_frames = input.size() / num_channels_;
size_t output_frames = output.size() / num_channels_;
int half_taps = kTapsPerFilter / 2, i;
while (output_frames > 0) {
if (output_offset_ >= input_index_ - half_taps) {
if (input_frames > 0) {
if (input_index_ == channel_buffer_size_) {
for (i = 0; i < num_channels_; ++i) {
memmove(channel_buffers_[i],
channel_buffers_[i] + channel_buffer_size_ - kTapsPerFilter,
kTapsPerFilter * sizeof(float));
}
output_offset_ -= channel_buffer_size_ - kTapsPerFilter;
input_index_ -= channel_buffer_size_ - kTapsPerFilter;
}
for (i = 0; i < num_channels_; ++i) {
channel_buffers_[i][input_index_] =
sample::ToFloat(input[samples_used++]);
}
input_index_++;
input_frames--;
} else
break;
} else {
for (i = 0; i < num_channels_; i++) {
output[samples_produced++] = sample::FromFloat(Subsample(i));
}
output_offset_ += (1.0f / factor_);
}
}
return {samples_used, samples_produced};
}
auto InitFilter(int index) -> void {
const double a0 = 0.35875;
const double a1 = 0.48829;
const double a2 = 0.14128;
const double a3 = 0.01168;
double fraction =
static_cast<double>(index) / static_cast<double>(kNumFilters);
double filter_sum = 0.0;
// "dist" is the absolute distance from the sinc maximum to the filter tap to
// be calculated, in radians "ratio" is that distance divided by half the tap
// count such that it reaches π at the window extremes
// Note that with this scaling, the odd terms of the Blackman-Harris
// calculation appear to be negated with respect to the reference formula
// version.
Filter& filter = sFilters[index];
std::array<double, kTapsPerFilter> working_buffer{};
for (int i = 0; i < kTapsPerFilter; ++i) {
double dist = fabs((kTapsPerFilter / 2.0 - 1.0) + fraction - i) * M_PI;
double ratio = dist / (kTapsPerFilter / 2.0);
double value;
if (dist != 0.0) {
value = sin(dist * kLowPassRatio) / (dist * kLowPassRatio);
// Blackman-Harris window
value *= a0 + a1 * cos(ratio) + a2 * cos(2 * ratio) + a3 * cos(3 * ratio);
} else {
value = 1.0;
}
working_buffer[i] = value;
filter_sum += value;
}
// filter should have unity DC gain
double scaler = 1.0 / filter_sum;
double error = 0.0;
for (int i = kTapsPerFilter / 2; i < kTapsPerFilter;
i = kTapsPerFilter - i - (i >= kTapsPerFilter / 2)) {
working_buffer[i] *= scaler;
filter[i] = working_buffer[i] - error;
error += static_cast<double>(filter[i]) - working_buffer[i];
}
}
auto Resampler::Subsample(int channel) -> float {
float sum1, sum2;
cpp::span<float> source{channel_buffers_[channel], channel_buffer_size_};
int offset_integral = std::floor(output_offset_);
source = source.subspan(offset_integral);
float offset_fractional = output_offset_ - offset_integral;
int filter_index = offset_fractional * kNumFilters;
offset_fractional *= kNumFilters;
sum1 = ApplyFilter(sFilters[filter_index],
{source.data() - kTapsPerFilter / 2 + 1, kTapsPerFilter});
offset_fractional -= filter_index;
sum2 = ApplyFilter(sFilters[filter_index + 1],
{source.data() - kTapsPerFilter / 2 + 1, kTapsPerFilter});
return (sum2 * offset_fractional) + (sum1 * (1.0f - offset_fractional));
}
auto Resampler::ApplyFilter(cpp::span<float> filter, cpp::span<float> input)
-> float {
float sum = 0.0;
for (int i = 0; i < kTapsPerFilter; i++) {
sum += filter[i] * input[i];
}
return sum;
}
} // namespace audio
|
