1 /*
2 * Copyright (c) 2012 The WebRTC project authors. All Rights Reserved.
3 *
4 * Use of this source code is governed by a BSD-style license
5 * that can be found in the LICENSE file in the root of the source
6 * tree. An additional intellectual property rights grant can be found
7 * in the file PATENTS. All contributing project authors may
8 * be found in the AUTHORS file in the root of the source tree.
9 */
10
11 #include "modules/audio_coding/neteq/normal.h"
12
13 #include <string.h> // memset, memcpy
14
15 #include <algorithm> // min
16
17 #include "api/audio_codecs/audio_decoder.h"
18 #include "common_audio/signal_processing/include/signal_processing_library.h"
19 #include "modules/audio_coding/neteq/audio_multi_vector.h"
20 #include "modules/audio_coding/neteq/background_noise.h"
21 #include "modules/audio_coding/neteq/decoder_database.h"
22 #include "modules/audio_coding/neteq/expand.h"
23 #include "rtc_base/checks.h"
24
25 namespace webrtc {
26
Process(const int16_t * input,size_t length,NetEq::Mode last_mode,AudioMultiVector * output)27 int Normal::Process(const int16_t* input,
28 size_t length,
29 NetEq::Mode last_mode,
30 AudioMultiVector* output) {
31 if (length == 0) {
32 // Nothing to process.
33 output->Clear();
34 return static_cast<int>(length);
35 }
36
37 RTC_DCHECK(output->Empty());
38 // Output should be empty at this point.
39 if (length % output->Channels() != 0) {
40 // The length does not match the number of channels.
41 output->Clear();
42 return 0;
43 }
44 output->PushBackInterleaved(rtc::ArrayView<const int16_t>(input, length));
45
46 const int fs_mult = fs_hz_ / 8000;
47 RTC_DCHECK_GT(fs_mult, 0);
48 // fs_shift = log2(fs_mult), rounded down.
49 // Note that |fs_shift| is not "exact" for 48 kHz.
50 // TODO(hlundin): Investigate this further.
51 const int fs_shift = 30 - WebRtcSpl_NormW32(fs_mult);
52
53 // Check if last RecOut call resulted in an Expand. If so, we have to take
54 // care of some cross-fading and unmuting.
55 if (last_mode == NetEq::Mode::kExpand) {
56 // Generate interpolation data using Expand.
57 // First, set Expand parameters to appropriate values.
58 expand_->SetParametersForNormalAfterExpand();
59
60 // Call Expand.
61 AudioMultiVector expanded(output->Channels());
62 expand_->Process(&expanded);
63 expand_->Reset();
64
65 size_t length_per_channel = length / output->Channels();
66 std::unique_ptr<int16_t[]> signal(new int16_t[length_per_channel]);
67 for (size_t channel_ix = 0; channel_ix < output->Channels(); ++channel_ix) {
68 // Set muting factor to the same as expand muting factor.
69 int16_t mute_factor = expand_->MuteFactor(channel_ix);
70
71 (*output)[channel_ix].CopyTo(length_per_channel, 0, signal.get());
72
73 // Find largest absolute value in new data.
74 int16_t decoded_max =
75 WebRtcSpl_MaxAbsValueW16(signal.get(), length_per_channel);
76 // Adjust muting factor if needed (to BGN level).
77 size_t energy_length =
78 std::min(static_cast<size_t>(fs_mult * 64), length_per_channel);
79 int scaling = 6 + fs_shift - WebRtcSpl_NormW32(decoded_max * decoded_max);
80 scaling = std::max(scaling, 0); // |scaling| should always be >= 0.
81 int32_t energy = WebRtcSpl_DotProductWithScale(signal.get(), signal.get(),
82 energy_length, scaling);
83 int32_t scaled_energy_length =
84 static_cast<int32_t>(energy_length >> scaling);
85 if (scaled_energy_length > 0) {
86 energy = energy / scaled_energy_length;
87 } else {
88 energy = 0;
89 }
90
91 int local_mute_factor = 16384; // 1.0 in Q14.
92 if ((energy != 0) && (energy > background_noise_.Energy(channel_ix))) {
93 // Normalize new frame energy to 15 bits.
94 scaling = WebRtcSpl_NormW32(energy) - 16;
95 // We want background_noise_.energy() / energy in Q14.
96 int32_t bgn_energy = WEBRTC_SPL_SHIFT_W32(
97 background_noise_.Energy(channel_ix), scaling + 14);
98 int16_t energy_scaled =
99 static_cast<int16_t>(WEBRTC_SPL_SHIFT_W32(energy, scaling));
100 int32_t ratio = WebRtcSpl_DivW32W16(bgn_energy, energy_scaled);
101 local_mute_factor =
102 std::min(local_mute_factor, WebRtcSpl_SqrtFloor(ratio << 14));
103 }
104 mute_factor = std::max<int16_t>(mute_factor, local_mute_factor);
105 RTC_DCHECK_LE(mute_factor, 16384);
106 RTC_DCHECK_GE(mute_factor, 0);
107
108 // If muted increase by 0.64 for every 20 ms (NB/WB 0.0040/0.0020 in Q14),
109 // or as fast as it takes to come back to full gain within the frame
110 // length.
111 const int back_to_fullscale_inc =
112 static_cast<int>((16384 - mute_factor) / length_per_channel);
113 const int increment = std::max(64 / fs_mult, back_to_fullscale_inc);
114 for (size_t i = 0; i < length_per_channel; i++) {
115 // Scale with mute factor.
116 RTC_DCHECK_LT(channel_ix, output->Channels());
117 RTC_DCHECK_LT(i, output->Size());
118 int32_t scaled_signal = (*output)[channel_ix][i] * mute_factor;
119 // Shift 14 with proper rounding.
120 (*output)[channel_ix][i] =
121 static_cast<int16_t>((scaled_signal + 8192) >> 14);
122 // Increase mute_factor towards 16384.
123 mute_factor =
124 static_cast<int16_t>(std::min(mute_factor + increment, 16384));
125 }
126
127 // Interpolate the expanded data into the new vector.
128 // (NB/WB/SWB32/SWB48 8/16/32/48 samples.)
129 size_t win_length = samples_per_ms_;
130 int16_t win_slope_Q14 = default_win_slope_Q14_;
131 RTC_DCHECK_LT(channel_ix, output->Channels());
132 if (win_length > output->Size()) {
133 win_length = output->Size();
134 win_slope_Q14 = (1 << 14) / static_cast<int16_t>(win_length);
135 }
136 int16_t win_up_Q14 = 0;
137 for (size_t i = 0; i < win_length; i++) {
138 win_up_Q14 += win_slope_Q14;
139 (*output)[channel_ix][i] =
140 (win_up_Q14 * (*output)[channel_ix][i] +
141 ((1 << 14) - win_up_Q14) * expanded[channel_ix][i] + (1 << 13)) >>
142 14;
143 }
144 RTC_DCHECK_GT(win_up_Q14,
145 (1 << 14) - 32); // Worst case rouding is a length of 34
146 }
147 } else if (last_mode == NetEq::Mode::kRfc3389Cng) {
148 RTC_DCHECK_EQ(output->Channels(), 1); // Not adapted for multi-channel yet.
149 static const size_t kCngLength = 48;
150 RTC_DCHECK_LE(8 * fs_mult, kCngLength);
151 int16_t cng_output[kCngLength];
152 ComfortNoiseDecoder* cng_decoder = decoder_database_->GetActiveCngDecoder();
153
154 if (cng_decoder) {
155 // Generate long enough for 48kHz.
156 if (!cng_decoder->Generate(cng_output, 0)) {
157 // Error returned; set return vector to all zeros.
158 memset(cng_output, 0, sizeof(cng_output));
159 }
160 } else {
161 // If no CNG instance is defined, just copy from the decoded data.
162 // (This will result in interpolating the decoded with itself.)
163 (*output)[0].CopyTo(fs_mult * 8, 0, cng_output);
164 }
165 // Interpolate the CNG into the new vector.
166 // (NB/WB/SWB32/SWB48 8/16/32/48 samples.)
167 size_t win_length = samples_per_ms_;
168 int16_t win_slope_Q14 = default_win_slope_Q14_;
169 if (win_length > kCngLength) {
170 win_length = kCngLength;
171 win_slope_Q14 = (1 << 14) / static_cast<int16_t>(win_length);
172 }
173 int16_t win_up_Q14 = 0;
174 for (size_t i = 0; i < win_length; i++) {
175 win_up_Q14 += win_slope_Q14;
176 (*output)[0][i] =
177 (win_up_Q14 * (*output)[0][i] +
178 ((1 << 14) - win_up_Q14) * cng_output[i] + (1 << 13)) >>
179 14;
180 }
181 RTC_DCHECK_GT(win_up_Q14,
182 (1 << 14) - 32); // Worst case rouding is a length of 34
183 }
184
185 return static_cast<int>(length);
186 }
187
188 } // namespace webrtc
189