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 25 namespace webrtc { 26 27 int 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 output->PushBackInterleaved(input, length); 41 int16_t* signal = &(*output)[0][0]; 42 43 const unsigned fs_mult = fs_hz_ / 8000; 44 assert(fs_mult > 0); 45 // fs_shift = log2(fs_mult), rounded down. 46 // Note that |fs_shift| is not "exact" for 48 kHz. 47 // TODO(hlundin): Investigate this further. 48 const int fs_shift = 30 - WebRtcSpl_NormW32(fs_mult); 49 50 // Check if last RecOut call resulted in an Expand. If so, we have to take 51 // care of some cross-fading and unmuting. 52 if (last_mode == kModeExpand) { 53 // Generate interpolation data using Expand. 54 // First, set Expand parameters to appropriate values. 55 expand_->SetParametersForNormalAfterExpand(); 56 57 // Call Expand. 58 AudioMultiVector expanded(output->Channels()); 59 expand_->Process(&expanded); 60 expand_->Reset(); 61 62 for (size_t channel_ix = 0; channel_ix < output->Channels(); ++channel_ix) { 63 // Adjust muting factor (main muting factor times expand muting factor). 64 external_mute_factor_array[channel_ix] = static_cast<int16_t>( 65 WEBRTC_SPL_MUL_16_16_RSFT(external_mute_factor_array[channel_ix], 66 expand_->MuteFactor(channel_ix), 14)); 67 68 int16_t* signal = &(*output)[channel_ix][0]; 69 size_t length_per_channel = length / output->Channels(); 70 // Find largest absolute value in new data. 71 int16_t decoded_max = WebRtcSpl_MaxAbsValueW16( 72 signal, static_cast<int>(length_per_channel)); 73 // Adjust muting factor if needed (to BGN level). 74 int energy_length = std::min(static_cast<int>(fs_mult * 64), 75 static_cast<int>(length_per_channel)); 76 int scaling = 6 + fs_shift 77 - WebRtcSpl_NormW32(decoded_max * decoded_max); 78 scaling = std::max(scaling, 0); // |scaling| should always be >= 0. 79 int32_t energy = WebRtcSpl_DotProductWithScale(signal, signal, 80 energy_length, scaling); 81 energy = energy / (energy_length >> scaling); 82 83 int mute_factor; 84 if ((energy != 0) && 85 (energy > background_noise_.Energy(channel_ix))) { 86 // Normalize new frame energy to 15 bits. 87 scaling = WebRtcSpl_NormW32(energy) - 16; 88 // We want background_noise_.energy() / energy in Q14. 89 int32_t bgn_energy = 90 background_noise_.Energy(channel_ix) << (scaling+14); 91 int16_t energy_scaled = energy << scaling; 92 int16_t ratio = WebRtcSpl_DivW32W16(bgn_energy, energy_scaled); 93 mute_factor = WebRtcSpl_SqrtFloor(static_cast<int32_t>(ratio) << 14); 94 } else { 95 mute_factor = 16384; // 1.0 in Q14. 96 } 97 if (mute_factor > external_mute_factor_array[channel_ix]) { 98 external_mute_factor_array[channel_ix] = std::min(mute_factor, 16384); 99 } 100 101 // If muted increase by 0.64 for every 20 ms (NB/WB 0.0040/0.0020 in Q14). 102 int16_t increment = 64 / fs_mult; 103 for (size_t i = 0; i < length_per_channel; i++) { 104 // Scale with mute factor. 105 assert(channel_ix < output->Channels()); 106 assert(i < output->Size()); 107 int32_t scaled_signal = (*output)[channel_ix][i] * 108 external_mute_factor_array[channel_ix]; 109 // Shift 14 with proper rounding. 110 (*output)[channel_ix][i] = (scaled_signal + 8192) >> 14; 111 // Increase mute_factor towards 16384. 112 external_mute_factor_array[channel_ix] = 113 std::min(external_mute_factor_array[channel_ix] + increment, 16384); 114 } 115 116 // Interpolate the expanded data into the new vector. 117 // (NB/WB/SWB32/SWB48 8/16/32/48 samples.) 118 assert(fs_shift < 3); // Will always be 0, 1, or, 2. 119 increment = 4 >> fs_shift; 120 int fraction = increment; 121 for (size_t i = 0; i < 8 * fs_mult; i++) { 122 // TODO(hlundin): Add 16 instead of 8 for correct rounding. Keeping 8 123 // now for legacy bit-exactness. 124 assert(channel_ix < output->Channels()); 125 assert(i < output->Size()); 126 (*output)[channel_ix][i] = 127 (fraction * (*output)[channel_ix][i] + 128 (32 - fraction) * expanded[channel_ix][i] + 8) >> 5; 129 fraction += increment; 130 } 131 } 132 } else if (last_mode == kModeRfc3389Cng) { 133 assert(output->Channels() == 1); // Not adapted for multi-channel yet. 134 static const int kCngLength = 32; 135 int16_t cng_output[kCngLength]; 136 // Reset mute factor and start up fresh. 137 external_mute_factor_array[0] = 16384; 138 AudioDecoder* cng_decoder = decoder_database_->GetActiveCngDecoder(); 139 140 if (cng_decoder) { 141 CNG_dec_inst* cng_inst = static_cast<CNG_dec_inst*>(cng_decoder->state()); 142 // Generate long enough for 32kHz. 143 if (WebRtcCng_Generate(cng_inst, cng_output, kCngLength, 0) < 0) { 144 // Error returned; set return vector to all zeros. 145 memset(cng_output, 0, sizeof(cng_output)); 146 } 147 } else { 148 // If no CNG instance is defined, just copy from the decoded data. 149 // (This will result in interpolating the decoded with itself.) 150 memcpy(cng_output, signal, fs_mult * 8 * sizeof(int16_t)); 151 } 152 // Interpolate the CNG into the new vector. 153 // (NB/WB/SWB32/SWB48 8/16/32/48 samples.) 154 assert(fs_shift < 3); // Will always be 0, 1, or, 2. 155 int16_t increment = 4 >> fs_shift; 156 int16_t fraction = increment; 157 for (size_t i = 0; i < 8 * fs_mult; i++) { 158 // TODO(hlundin): Add 16 instead of 8 for correct rounding. Keeping 8 now 159 // for legacy bit-exactness. 160 signal[i] = 161 (fraction * signal[i] + (32 - fraction) * cng_output[i] + 8) >> 5; 162 fraction += increment; 163 } 164 } else if (external_mute_factor_array[0] < 16384) { 165 // Previous was neither of Expand, FadeToBGN or RFC3389_CNG, but we are 166 // still ramping up from previous muting. 167 // If muted increase by 0.64 for every 20 ms (NB/WB 0.0040/0.0020 in Q14). 168 int16_t increment = 64 / fs_mult; 169 size_t length_per_channel = length / output->Channels(); 170 for (size_t i = 0; i < length_per_channel; i++) { 171 for (size_t channel_ix = 0; channel_ix < output->Channels(); 172 ++channel_ix) { 173 // Scale with mute factor. 174 assert(channel_ix < output->Channels()); 175 assert(i < output->Size()); 176 int32_t scaled_signal = (*output)[channel_ix][i] * 177 external_mute_factor_array[channel_ix]; 178 // Shift 14 with proper rounding. 179 (*output)[channel_ix][i] = (scaled_signal + 8192) >> 14; 180 // Increase mute_factor towards 16384. 181 external_mute_factor_array[channel_ix] = 182 std::min(16384, external_mute_factor_array[channel_ix] + increment); 183 } 184 } 185 } 186 187 return static_cast<int>(length); 188 } 189 190 } // namespace webrtc 191