henrik.lundin@webrtc.org | d94659d | 2013-01-29 12:09:21 +0000 | [diff] [blame^] | 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/neteq4/normal.h" |
| 12 | |
| 13 | #include <algorithm> // min |
| 14 | #include <cstring> // memset, memcpy |
| 15 | |
| 16 | #include "webrtc/common_audio/signal_processing/include/signal_processing_library.h" |
| 17 | #include "webrtc/modules/audio_coding/codecs/cng/include/webrtc_cng.h" |
| 18 | #include "webrtc/modules/audio_coding/neteq4/audio_multi_vector.h" |
| 19 | #include "webrtc/modules/audio_coding/neteq4/background_noise.h" |
| 20 | #include "webrtc/modules/audio_coding/neteq4/decoder_database.h" |
| 21 | #include "webrtc/modules/audio_coding/neteq4/expand.h" |
| 22 | #include "webrtc/modules/audio_coding/neteq4/interface/audio_decoder.h" |
| 23 | |
| 24 | namespace webrtc { |
| 25 | |
| 26 | int Normal::Process(const int16_t* input, |
| 27 | size_t length, |
| 28 | Modes last_mode, |
| 29 | int16_t* external_mute_factor_array, |
| 30 | AudioMultiVector<int16_t>* output) { |
| 31 | if (length == 0) { |
| 32 | // Nothing to process. |
| 33 | output->Clear(); |
| 34 | return length; |
| 35 | } |
| 36 | |
| 37 | assert(output->Empty()); |
| 38 | // Output should be empty at this point. |
| 39 | output->PushBackInterleaved(input, length); |
| 40 | int16_t* signal = &(*output)[0][0]; |
| 41 | |
| 42 | const unsigned fs_mult = fs_hz_ / 8000; |
| 43 | assert(fs_mult > 0); |
| 44 | // fs_shift = log2(fs_mult), rounded down. |
| 45 | // Note that |fs_shift| is not "exact" for 48 kHz. |
| 46 | // TODO(hlundin): Investigate this further. |
| 47 | const int fs_shift = 30 - WebRtcSpl_NormW32(fs_mult); |
| 48 | |
| 49 | // Check if last RecOut call resulted in an Expand. If so, we have to take |
| 50 | // care of some cross-fading and unmuting. |
| 51 | if (last_mode == kModeExpand) { |
| 52 | // Generate interpolation data using Expand. |
| 53 | // First, set Expand parameters to appropriate values. |
| 54 | expand_->SetParametersForNormalAfterExpand(); |
| 55 | |
| 56 | // Call Expand. |
| 57 | AudioMultiVector<int16_t> expanded(output->Channels()); |
| 58 | expand_->Process(&expanded); |
| 59 | expand_->Reset(); |
| 60 | |
| 61 | for (size_t channel_ix = 0; channel_ix < output->Channels(); ++channel_ix) { |
| 62 | // Adjust muting factor (main muting factor times expand muting factor). |
| 63 | external_mute_factor_array[channel_ix] = static_cast<int16_t>( |
| 64 | WEBRTC_SPL_MUL_16_16_RSFT(external_mute_factor_array[channel_ix], |
| 65 | expand_->MuteFactor(channel_ix), 14)); |
| 66 | |
| 67 | int16_t* signal = &(*output)[channel_ix][0]; |
| 68 | size_t length_per_channel = length / output->Channels(); |
| 69 | // Find largest absolute value in new data. |
| 70 | int16_t decoded_max = WebRtcSpl_MaxAbsValueW16(signal, |
| 71 | length_per_channel); |
| 72 | // Adjust muting factor if needed (to BGN level). |
| 73 | int energy_length = std::min(static_cast<size_t>(fs_mult * 64), |
| 74 | length_per_channel); |
| 75 | int scaling = 6 + fs_shift |
| 76 | - WebRtcSpl_NormW32(decoded_max * decoded_max); |
| 77 | scaling = std::max(scaling, 0); // |scaling| should always be >= 0. |
| 78 | int32_t energy = WebRtcSpl_DotProductWithScale(signal, signal, |
| 79 | energy_length, scaling); |
| 80 | energy = energy / (energy_length >> scaling); |
| 81 | |
| 82 | int mute_factor; |
| 83 | if ((energy != 0) && |
| 84 | (energy > background_noise_.Energy(channel_ix))) { |
| 85 | // Normalize new frame energy to 15 bits. |
| 86 | scaling = WebRtcSpl_NormW32(energy) - 16; |
| 87 | // We want background_noise_.energy() / energy in Q14. |
| 88 | int32_t bgn_energy = |
| 89 | background_noise_.Energy(channel_ix) << (scaling+14); |
| 90 | int16_t energy_scaled = energy << scaling; |
| 91 | int16_t ratio = WebRtcSpl_DivW32W16(bgn_energy, energy_scaled); |
| 92 | mute_factor = WebRtcSpl_SqrtFloor(static_cast<int32_t>(ratio) << 14); |
| 93 | } else { |
| 94 | mute_factor = 16384; // 1.0 in Q14. |
| 95 | } |
| 96 | if (mute_factor > external_mute_factor_array[channel_ix]) { |
| 97 | external_mute_factor_array[channel_ix] = std::min(mute_factor, 16384); |
| 98 | } |
| 99 | |
| 100 | // If muted increase by 0.64 for every 20 ms (NB/WB 0.0040/0.0020 in Q14). |
| 101 | int16_t increment = 64 / fs_mult; |
| 102 | for (size_t i = 0; i < length_per_channel; i++) { |
| 103 | // Scale with mute factor. |
| 104 | assert(channel_ix < output->Channels()); |
| 105 | assert(i < output->Size()); |
| 106 | int32_t scaled_signal = (*output)[channel_ix][i] * |
| 107 | external_mute_factor_array[channel_ix]; |
| 108 | // Shift 14 with proper rounding. |
| 109 | (*output)[channel_ix][i] = (scaled_signal + 8192) >> 14; |
| 110 | // Increase mute_factor towards 16384. |
| 111 | external_mute_factor_array[channel_ix] = |
| 112 | std::min(external_mute_factor_array[channel_ix] + increment, 16384); |
| 113 | } |
| 114 | |
| 115 | // Interpolate the expanded data into the new vector. |
| 116 | // (NB/WB/SWB32/SWB48 8/16/32/48 samples.) |
| 117 | assert(fs_shift < 3); // Will always be 0, 1, or, 2. |
| 118 | increment = 4 >> fs_shift; |
| 119 | int fraction = increment; |
| 120 | for (size_t i = 0; i < 8 * fs_mult; i++) { |
| 121 | // TODO(hlundin): Add 16 instead of 8 for correct rounding. Keeping 8 |
| 122 | // now for legacy bit-exactness. |
| 123 | assert(channel_ix < output->Channels()); |
| 124 | assert(i < output->Size()); |
| 125 | (*output)[channel_ix][i] = |
| 126 | (fraction * (*output)[channel_ix][i] + |
| 127 | (32 - fraction) * expanded[channel_ix][i] + 8) >> 5; |
| 128 | fraction += increment; |
| 129 | } |
| 130 | } |
| 131 | } else if (last_mode == kModeRfc3389Cng) { |
| 132 | assert(output->Channels() == 1); // Not adapted for multi-channel yet. |
| 133 | static const int kCngLength = 32; |
| 134 | int16_t cng_output[kCngLength]; |
| 135 | // Reset mute factor and start up fresh. |
| 136 | external_mute_factor_array[0] = 16384; |
| 137 | AudioDecoder* cng_decoder = decoder_database_->GetActiveCngDecoder(); |
| 138 | |
| 139 | if (cng_decoder) { |
| 140 | CNG_dec_inst* cng_inst = static_cast<CNG_dec_inst*>(cng_decoder->state()); |
| 141 | // Generate long enough for 32kHz. |
| 142 | if (WebRtcCng_Generate(cng_inst, cng_output, kCngLength, 0) < 0) { |
| 143 | // Error returned; set return vector to all zeros. |
| 144 | memset(cng_output, 0, sizeof(cng_output)); |
| 145 | } |
| 146 | } else { |
| 147 | // If no CNG instance is defined, just copy from the decoded data. |
| 148 | // (This will result in interpolating the decoded with itself.) |
| 149 | memcpy(cng_output, signal, fs_mult * 8 * sizeof(int16_t)); |
| 150 | } |
| 151 | // Interpolate the CNG into the new vector. |
| 152 | // (NB/WB/SWB32/SWB48 8/16/32/48 samples.) |
| 153 | assert(fs_shift < 3); // Will always be 0, 1, or, 2. |
| 154 | int16_t increment = 4 >> fs_shift; |
| 155 | int16_t fraction = increment; |
| 156 | for (size_t i = 0; i < 8 * fs_mult; i++) { |
| 157 | // TODO(hlundin): Add 16 instead of 8 for correct rounding. Keeping 8 now |
| 158 | // for legacy bit-exactness. |
| 159 | signal[i] = |
| 160 | (fraction * signal[i] + (32 - fraction) * cng_output[i] + 8) >> 5; |
| 161 | fraction += increment; |
| 162 | } |
| 163 | } else if (external_mute_factor_array[0] < 16384) { |
| 164 | // Previous was neither of Expand, FadeToBGN or RFC3389_CNG, but we are |
| 165 | // still ramping up from previous muting. |
| 166 | // If muted increase by 0.64 for every 20 ms (NB/WB 0.0040/0.0020 in Q14). |
| 167 | int16_t increment = 64 / fs_mult; |
| 168 | size_t length_per_channel = length / output->Channels(); |
| 169 | for (size_t i = 0; i < length_per_channel; i++) { |
| 170 | for (size_t channel_ix = 0; channel_ix < output->Channels(); |
| 171 | ++channel_ix) { |
| 172 | // Scale with mute factor. |
| 173 | assert(channel_ix < output->Channels()); |
| 174 | assert(i < output->Size()); |
| 175 | int32_t scaled_signal = (*output)[channel_ix][i] * |
| 176 | external_mute_factor_array[channel_ix]; |
| 177 | // Shift 14 with proper rounding. |
| 178 | (*output)[channel_ix][i] = (scaled_signal + 8192) >> 14; |
| 179 | // Increase mute_factor towards 16384. |
| 180 | external_mute_factor_array[channel_ix] = |
| 181 | std::min(16384, external_mute_factor_array[channel_ix] + increment); |
| 182 | } |
| 183 | } |
| 184 | } |
| 185 | |
| 186 | return length; |
| 187 | } |
| 188 | |
| 189 | } // namespace webrtc |