normal.cc revision ee0fb187a583b0c66b2c9fc8571411dca510ce7b
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 "webrtc/modules/audio_coding/neteq/normal.h" 12 13#include <string.h> // memset, memcpy 14 15#include <algorithm> // min 16 17#include "webrtc/common_audio/signal_processing/include/signal_processing_library.h" 18#include "webrtc/modules/audio_coding/codecs/cng/include/webrtc_cng.h" 19#include "webrtc/modules/audio_coding/neteq/audio_multi_vector.h" 20#include "webrtc/modules/audio_coding/neteq/background_noise.h" 21#include "webrtc/modules/audio_coding/neteq/decoder_database.h" 22#include "webrtc/modules/audio_coding/neteq/expand.h" 23#include "webrtc/modules/audio_coding/neteq/interface/audio_decoder.h" 24 25namespace webrtc { 26 27int Normal::Process(const int16_t* input, 28 size_t length, 29 Modes last_mode, 30 int16_t* external_mute_factor_array, 31 AudioMultiVector* output) { 32 if (length == 0) { 33 // Nothing to process. 34 output->Clear(); 35 return static_cast<int>(length); 36 } 37 38 assert(output->Empty()); 39 // Output should be empty at this point. 40 if (length % output->Channels() != 0) { 41 // The length does not match the number of channels. 42 output->Clear(); 43 return 0; 44 } 45 output->PushBackInterleaved(input, length); 46 int16_t* signal = &(*output)[0][0]; 47 48 const unsigned fs_mult = fs_hz_ / 8000; 49 assert(fs_mult > 0); 50 // fs_shift = log2(fs_mult), rounded down. 51 // Note that |fs_shift| is not "exact" for 48 kHz. 52 // TODO(hlundin): Investigate this further. 53 const int fs_shift = 30 - WebRtcSpl_NormW32(fs_mult); 54 55 // Check if last RecOut call resulted in an Expand. If so, we have to take 56 // care of some cross-fading and unmuting. 57 if (last_mode == kModeExpand) { 58 // Generate interpolation data using Expand. 59 // First, set Expand parameters to appropriate values. 60 expand_->SetParametersForNormalAfterExpand(); 61 62 // Call Expand. 63 AudioMultiVector expanded(output->Channels()); 64 expand_->Process(&expanded); 65 expand_->Reset(); 66 67 for (size_t channel_ix = 0; channel_ix < output->Channels(); ++channel_ix) { 68 // Adjust muting factor (main muting factor times expand muting factor). 69 external_mute_factor_array[channel_ix] = static_cast<int16_t>( 70 WEBRTC_SPL_MUL_16_16_RSFT(external_mute_factor_array[channel_ix], 71 expand_->MuteFactor(channel_ix), 14)); 72 73 int16_t* signal = &(*output)[channel_ix][0]; 74 size_t length_per_channel = length / output->Channels(); 75 // Find largest absolute value in new data. 76 int16_t decoded_max = WebRtcSpl_MaxAbsValueW16( 77 signal, static_cast<int>(length_per_channel)); 78 // Adjust muting factor if needed (to BGN level). 79 int energy_length = std::min(static_cast<int>(fs_mult * 64), 80 static_cast<int>(length_per_channel)); 81 int scaling = 6 + fs_shift 82 - WebRtcSpl_NormW32(decoded_max * decoded_max); 83 scaling = std::max(scaling, 0); // |scaling| should always be >= 0. 84 int32_t energy = WebRtcSpl_DotProductWithScale(signal, signal, 85 energy_length, scaling); 86 if ((energy_length >> scaling) > 0) { 87 energy = energy / (energy_length >> scaling); 88 } else { 89 energy = 0; 90 } 91 92 int mute_factor; 93 if ((energy != 0) && 94 (energy > background_noise_.Energy(channel_ix))) { 95 // Normalize new frame energy to 15 bits. 96 scaling = WebRtcSpl_NormW32(energy) - 16; 97 // We want background_noise_.energy() / energy in Q14. 98 int32_t bgn_energy = 99 background_noise_.Energy(channel_ix) << (scaling+14); 100 int16_t energy_scaled = energy << scaling; 101 int16_t ratio = WebRtcSpl_DivW32W16(bgn_energy, energy_scaled); 102 mute_factor = WebRtcSpl_SqrtFloor(static_cast<int32_t>(ratio) << 14); 103 } else { 104 mute_factor = 16384; // 1.0 in Q14. 105 } 106 if (mute_factor > external_mute_factor_array[channel_ix]) { 107 external_mute_factor_array[channel_ix] = std::min(mute_factor, 16384); 108 } 109 110 // If muted increase by 0.64 for every 20 ms (NB/WB 0.0040/0.0020 in Q14). 111 int16_t increment = 64 / fs_mult; 112 for (size_t i = 0; i < length_per_channel; i++) { 113 // Scale with mute factor. 114 assert(channel_ix < output->Channels()); 115 assert(i < output->Size()); 116 int32_t scaled_signal = (*output)[channel_ix][i] * 117 external_mute_factor_array[channel_ix]; 118 // Shift 14 with proper rounding. 119 (*output)[channel_ix][i] = (scaled_signal + 8192) >> 14; 120 // Increase mute_factor towards 16384. 121 external_mute_factor_array[channel_ix] = 122 std::min(external_mute_factor_array[channel_ix] + increment, 16384); 123 } 124 125 // Interpolate the expanded data into the new vector. 126 // (NB/WB/SWB32/SWB48 8/16/32/48 samples.) 127 assert(fs_shift < 3); // Will always be 0, 1, or, 2. 128 increment = 4 >> fs_shift; 129 int fraction = increment; 130 for (size_t i = 0; i < 8 * fs_mult; i++) { 131 // TODO(hlundin): Add 16 instead of 8 for correct rounding. Keeping 8 132 // now for legacy bit-exactness. 133 assert(channel_ix < output->Channels()); 134 assert(i < output->Size()); 135 (*output)[channel_ix][i] = 136 (fraction * (*output)[channel_ix][i] + 137 (32 - fraction) * expanded[channel_ix][i] + 8) >> 5; 138 fraction += increment; 139 } 140 } 141 } else if (last_mode == kModeRfc3389Cng) { 142 assert(output->Channels() == 1); // Not adapted for multi-channel yet. 143 static const int kCngLength = 32; 144 int16_t cng_output[kCngLength]; 145 // Reset mute factor and start up fresh. 146 external_mute_factor_array[0] = 16384; 147 AudioDecoder* cng_decoder = decoder_database_->GetActiveCngDecoder(); 148 149 if (cng_decoder) { 150 CNG_dec_inst* cng_inst = static_cast<CNG_dec_inst*>(cng_decoder->state()); 151 // Generate long enough for 32kHz. 152 if (WebRtcCng_Generate(cng_inst, cng_output, kCngLength, 0) < 0) { 153 // Error returned; set return vector to all zeros. 154 memset(cng_output, 0, sizeof(cng_output)); 155 } 156 } else { 157 // If no CNG instance is defined, just copy from the decoded data. 158 // (This will result in interpolating the decoded with itself.) 159 memcpy(cng_output, signal, fs_mult * 8 * sizeof(int16_t)); 160 } 161 // Interpolate the CNG into the new vector. 162 // (NB/WB/SWB32/SWB48 8/16/32/48 samples.) 163 assert(fs_shift < 3); // Will always be 0, 1, or, 2. 164 int16_t increment = 4 >> fs_shift; 165 int16_t fraction = increment; 166 for (size_t i = 0; i < 8 * fs_mult; i++) { 167 // TODO(hlundin): Add 16 instead of 8 for correct rounding. Keeping 8 now 168 // for legacy bit-exactness. 169 signal[i] = 170 (fraction * signal[i] + (32 - fraction) * cng_output[i] + 8) >> 5; 171 fraction += increment; 172 } 173 } else if (external_mute_factor_array[0] < 16384) { 174 // Previous was neither of Expand, FadeToBGN or RFC3389_CNG, but we are 175 // still ramping up from previous muting. 176 // If muted increase by 0.64 for every 20 ms (NB/WB 0.0040/0.0020 in Q14). 177 int16_t increment = 64 / fs_mult; 178 size_t length_per_channel = length / output->Channels(); 179 for (size_t i = 0; i < length_per_channel; i++) { 180 for (size_t channel_ix = 0; channel_ix < output->Channels(); 181 ++channel_ix) { 182 // Scale with mute factor. 183 assert(channel_ix < output->Channels()); 184 assert(i < output->Size()); 185 int32_t scaled_signal = (*output)[channel_ix][i] * 186 external_mute_factor_array[channel_ix]; 187 // Shift 14 with proper rounding. 188 (*output)[channel_ix][i] = (scaled_signal + 8192) >> 14; 189 // Increase mute_factor towards 16384. 190 external_mute_factor_array[channel_ix] = 191 std::min(16384, external_mute_factor_array[channel_ix] + increment); 192 } 193 } 194 } 195 196 return static_cast<int>(length); 197} 198 199} // namespace webrtc 200