niklase@google.com | 470e71d | 2011-07-07 08:21:25 +0000 | [diff] [blame] | 1 | /* |
bjornv@webrtc.org | a496b03 | 2012-03-20 12:53:06 +0000 | [diff] [blame] | 2 | * Copyright (c) 2012 The WebRTC project authors. All Rights Reserved. |
niklase@google.com | 470e71d | 2011-07-07 08:21:25 +0000 | [diff] [blame] | 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 | |
pbos@webrtc.org | aa30bb7 | 2013-05-27 09:49:58 +0000 | [diff] [blame] | 11 | #include "webrtc/common_audio/vad/vad_sp.h" |
bjornv@webrtc.org | 2111d3b | 2011-10-14 12:58:34 +0000 | [diff] [blame] | 12 | |
bjornv@webrtc.org | 226c5a1 | 2012-01-04 09:15:12 +0000 | [diff] [blame] | 13 | #include <assert.h> |
| 14 | |
pbos@webrtc.org | aa30bb7 | 2013-05-27 09:49:58 +0000 | [diff] [blame] | 15 | #include "webrtc/common_audio/signal_processing/include/signal_processing_library.h" |
| 16 | #include "webrtc/common_audio/vad/vad_core.h" |
| 17 | #include "webrtc/typedefs.h" |
bjornv@webrtc.org | 2111d3b | 2011-10-14 12:58:34 +0000 | [diff] [blame] | 18 | |
bjornv@webrtc.org | 226c5a1 | 2012-01-04 09:15:12 +0000 | [diff] [blame] | 19 | // Allpass filter coefficients, upper and lower, in Q13. |
| 20 | // Upper: 0.64, Lower: 0.17. |
bjornv@webrtc.org | a496b03 | 2012-03-20 12:53:06 +0000 | [diff] [blame] | 21 | static const int16_t kAllPassCoefsQ13[2] = { 5243, 1392 }; // Q13. |
| 22 | static const int16_t kSmoothingDown = 6553; // 0.2 in Q15. |
| 23 | static const int16_t kSmoothingUp = 32439; // 0.99 in Q15. |
niklase@google.com | 470e71d | 2011-07-07 08:21:25 +0000 | [diff] [blame] | 24 | |
bjornv@webrtc.org | 226c5a1 | 2012-01-04 09:15:12 +0000 | [diff] [blame] | 25 | // TODO(bjornv): Move this function to vad_filterbank.c. |
| 26 | // Downsampling filter based on splitting filter and allpass functions. |
andrew@webrtc.org | 65f9338 | 2014-04-30 16:44:13 +0000 | [diff] [blame] | 27 | void WebRtcVad_Downsampling(const int16_t* signal_in, |
bjornv@webrtc.org | 226c5a1 | 2012-01-04 09:15:12 +0000 | [diff] [blame] | 28 | int16_t* signal_out, |
| 29 | int32_t* filter_state, |
Peter Kasting | dce40cf | 2015-08-24 14:52:23 -0700 | [diff] [blame^] | 30 | size_t in_length) { |
bjornv@webrtc.org | 226c5a1 | 2012-01-04 09:15:12 +0000 | [diff] [blame] | 31 | int16_t tmp16_1 = 0, tmp16_2 = 0; |
| 32 | int32_t tmp32_1 = filter_state[0]; |
| 33 | int32_t tmp32_2 = filter_state[1]; |
Peter Kasting | dce40cf | 2015-08-24 14:52:23 -0700 | [diff] [blame^] | 34 | size_t n = 0; |
| 35 | // Downsampling by 2 gives half length. |
| 36 | size_t half_length = (in_length >> 1); |
niklase@google.com | 470e71d | 2011-07-07 08:21:25 +0000 | [diff] [blame] | 37 | |
bjornv@webrtc.org | 226c5a1 | 2012-01-04 09:15:12 +0000 | [diff] [blame] | 38 | // Filter coefficients in Q13, filter state in Q0. |
| 39 | for (n = 0; n < half_length; n++) { |
| 40 | // All-pass filtering upper branch. |
| 41 | tmp16_1 = (int16_t) ((tmp32_1 >> 1) + |
Bjorn Volcker | 3fbf99c | 2015-03-25 14:37:24 +0100 | [diff] [blame] | 42 | ((kAllPassCoefsQ13[0] * *signal_in) >> 14)); |
bjornv@webrtc.org | 226c5a1 | 2012-01-04 09:15:12 +0000 | [diff] [blame] | 43 | *signal_out = tmp16_1; |
Bjorn Volcker | 3fbf99c | 2015-03-25 14:37:24 +0100 | [diff] [blame] | 44 | tmp32_1 = (int32_t)(*signal_in++) - ((kAllPassCoefsQ13[0] * tmp16_1) >> 12); |
niklase@google.com | 470e71d | 2011-07-07 08:21:25 +0000 | [diff] [blame] | 45 | |
bjornv@webrtc.org | 226c5a1 | 2012-01-04 09:15:12 +0000 | [diff] [blame] | 46 | // All-pass filtering lower branch. |
| 47 | tmp16_2 = (int16_t) ((tmp32_2 >> 1) + |
Bjorn Volcker | 3fbf99c | 2015-03-25 14:37:24 +0100 | [diff] [blame] | 48 | ((kAllPassCoefsQ13[1] * *signal_in) >> 14)); |
bjornv@webrtc.org | 226c5a1 | 2012-01-04 09:15:12 +0000 | [diff] [blame] | 49 | *signal_out++ += tmp16_2; |
Bjorn Volcker | 3fbf99c | 2015-03-25 14:37:24 +0100 | [diff] [blame] | 50 | tmp32_2 = (int32_t)(*signal_in++) - ((kAllPassCoefsQ13[1] * tmp16_2) >> 12); |
bjornv@webrtc.org | 226c5a1 | 2012-01-04 09:15:12 +0000 | [diff] [blame] | 51 | } |
| 52 | // Store the filter states. |
| 53 | filter_state[0] = tmp32_1; |
| 54 | filter_state[1] = tmp32_2; |
niklase@google.com | 470e71d | 2011-07-07 08:21:25 +0000 | [diff] [blame] | 55 | } |
| 56 | |
bjornv@webrtc.org | 226c5a1 | 2012-01-04 09:15:12 +0000 | [diff] [blame] | 57 | // Inserts |feature_value| into |low_value_vector|, if it is one of the 16 |
| 58 | // smallest values the last 100 frames. Then calculates and returns the median |
| 59 | // of the five smallest values. |
| 60 | int16_t WebRtcVad_FindMinimum(VadInstT* self, |
| 61 | int16_t feature_value, |
| 62 | int channel) { |
| 63 | int i = 0, j = 0; |
| 64 | int position = -1; |
| 65 | // Offset to beginning of the 16 minimum values in memory. |
bjornv@webrtc.org | eec739f | 2012-06-11 07:57:57 +0000 | [diff] [blame] | 66 | const int offset = (channel << 4); |
bjornv@webrtc.org | 226c5a1 | 2012-01-04 09:15:12 +0000 | [diff] [blame] | 67 | int16_t current_median = 1600; |
| 68 | int16_t alpha = 0; |
| 69 | int32_t tmp32 = 0; |
| 70 | // Pointer to memory for the 16 minimum values and the age of each value of |
| 71 | // the |channel|. |
bjornv@webrtc.org | eec739f | 2012-06-11 07:57:57 +0000 | [diff] [blame] | 72 | int16_t* age = &self->index_vector[offset]; |
| 73 | int16_t* smallest_values = &self->low_value_vector[offset]; |
niklase@google.com | 470e71d | 2011-07-07 08:21:25 +0000 | [diff] [blame] | 74 | |
bjornv@webrtc.org | a496b03 | 2012-03-20 12:53:06 +0000 | [diff] [blame] | 75 | assert(channel < kNumChannels); |
niklase@google.com | 470e71d | 2011-07-07 08:21:25 +0000 | [diff] [blame] | 76 | |
bjornv@webrtc.org | eec739f | 2012-06-11 07:57:57 +0000 | [diff] [blame] | 77 | // Each value in |smallest_values| is getting 1 loop older. Update |age|, and |
| 78 | // remove old values. |
bjornv@webrtc.org | 226c5a1 | 2012-01-04 09:15:12 +0000 | [diff] [blame] | 79 | for (i = 0; i < 16; i++) { |
bjornv@webrtc.org | eec739f | 2012-06-11 07:57:57 +0000 | [diff] [blame] | 80 | if (age[i] != 100) { |
| 81 | age[i]++; |
bjornv@webrtc.org | 226c5a1 | 2012-01-04 09:15:12 +0000 | [diff] [blame] | 82 | } else { |
bjornv@webrtc.org | eec739f | 2012-06-11 07:57:57 +0000 | [diff] [blame] | 83 | // Too old value. Remove from memory and shift larger values downwards. |
bjornv@webrtc.org | 226c5a1 | 2012-01-04 09:15:12 +0000 | [diff] [blame] | 84 | for (j = i; j < 16; j++) { |
bjornv@webrtc.org | eec739f | 2012-06-11 07:57:57 +0000 | [diff] [blame] | 85 | smallest_values[j] = smallest_values[j + 1]; |
| 86 | age[j] = age[j + 1]; |
bjornv@webrtc.org | 226c5a1 | 2012-01-04 09:15:12 +0000 | [diff] [blame] | 87 | } |
bjornv@webrtc.org | eec739f | 2012-06-11 07:57:57 +0000 | [diff] [blame] | 88 | age[15] = 101; |
| 89 | smallest_values[15] = 10000; |
bjornv@webrtc.org | 226c5a1 | 2012-01-04 09:15:12 +0000 | [diff] [blame] | 90 | } |
| 91 | } |
niklase@google.com | 470e71d | 2011-07-07 08:21:25 +0000 | [diff] [blame] | 92 | |
bjornv@webrtc.org | 226c5a1 | 2012-01-04 09:15:12 +0000 | [diff] [blame] | 93 | // Check if |feature_value| is smaller than any of the values in |
bjornv@webrtc.org | eec739f | 2012-06-11 07:57:57 +0000 | [diff] [blame] | 94 | // |smallest_values|. If so, find the |position| where to insert the new value |
| 95 | // (|feature_value|). |
| 96 | if (feature_value < smallest_values[7]) { |
| 97 | if (feature_value < smallest_values[3]) { |
| 98 | if (feature_value < smallest_values[1]) { |
| 99 | if (feature_value < smallest_values[0]) { |
bjornv@webrtc.org | 226c5a1 | 2012-01-04 09:15:12 +0000 | [diff] [blame] | 100 | position = 0; |
| 101 | } else { |
| 102 | position = 1; |
niklase@google.com | 470e71d | 2011-07-07 08:21:25 +0000 | [diff] [blame] | 103 | } |
bjornv@webrtc.org | eec739f | 2012-06-11 07:57:57 +0000 | [diff] [blame] | 104 | } else if (feature_value < smallest_values[2]) { |
bjornv@webrtc.org | 226c5a1 | 2012-01-04 09:15:12 +0000 | [diff] [blame] | 105 | position = 2; |
| 106 | } else { |
| 107 | position = 3; |
| 108 | } |
bjornv@webrtc.org | eec739f | 2012-06-11 07:57:57 +0000 | [diff] [blame] | 109 | } else if (feature_value < smallest_values[5]) { |
| 110 | if (feature_value < smallest_values[4]) { |
bjornv@webrtc.org | 226c5a1 | 2012-01-04 09:15:12 +0000 | [diff] [blame] | 111 | position = 4; |
| 112 | } else { |
| 113 | position = 5; |
| 114 | } |
bjornv@webrtc.org | eec739f | 2012-06-11 07:57:57 +0000 | [diff] [blame] | 115 | } else if (feature_value < smallest_values[6]) { |
bjornv@webrtc.org | 226c5a1 | 2012-01-04 09:15:12 +0000 | [diff] [blame] | 116 | position = 6; |
| 117 | } else { |
| 118 | position = 7; |
niklase@google.com | 470e71d | 2011-07-07 08:21:25 +0000 | [diff] [blame] | 119 | } |
bjornv@webrtc.org | eec739f | 2012-06-11 07:57:57 +0000 | [diff] [blame] | 120 | } else if (feature_value < smallest_values[15]) { |
| 121 | if (feature_value < smallest_values[11]) { |
| 122 | if (feature_value < smallest_values[9]) { |
| 123 | if (feature_value < smallest_values[8]) { |
bjornv@webrtc.org | 226c5a1 | 2012-01-04 09:15:12 +0000 | [diff] [blame] | 124 | position = 8; |
| 125 | } else { |
| 126 | position = 9; |
niklase@google.com | 470e71d | 2011-07-07 08:21:25 +0000 | [diff] [blame] | 127 | } |
bjornv@webrtc.org | eec739f | 2012-06-11 07:57:57 +0000 | [diff] [blame] | 128 | } else if (feature_value < smallest_values[10]) { |
bjornv@webrtc.org | 226c5a1 | 2012-01-04 09:15:12 +0000 | [diff] [blame] | 129 | position = 10; |
| 130 | } else { |
| 131 | position = 11; |
| 132 | } |
bjornv@webrtc.org | eec739f | 2012-06-11 07:57:57 +0000 | [diff] [blame] | 133 | } else if (feature_value < smallest_values[13]) { |
| 134 | if (feature_value < smallest_values[12]) { |
bjornv@webrtc.org | 226c5a1 | 2012-01-04 09:15:12 +0000 | [diff] [blame] | 135 | position = 12; |
| 136 | } else { |
| 137 | position = 13; |
| 138 | } |
bjornv@webrtc.org | eec739f | 2012-06-11 07:57:57 +0000 | [diff] [blame] | 139 | } else if (feature_value < smallest_values[14]) { |
bjornv@webrtc.org | 226c5a1 | 2012-01-04 09:15:12 +0000 | [diff] [blame] | 140 | position = 14; |
| 141 | } else { |
| 142 | position = 15; |
niklase@google.com | 470e71d | 2011-07-07 08:21:25 +0000 | [diff] [blame] | 143 | } |
bjornv@webrtc.org | 226c5a1 | 2012-01-04 09:15:12 +0000 | [diff] [blame] | 144 | } |
niklase@google.com | 470e71d | 2011-07-07 08:21:25 +0000 | [diff] [blame] | 145 | |
bjornv@webrtc.org | eec739f | 2012-06-11 07:57:57 +0000 | [diff] [blame] | 146 | // If we have detected a new small value, insert it at the correct position |
| 147 | // and shift larger values up. |
bjornv@webrtc.org | 226c5a1 | 2012-01-04 09:15:12 +0000 | [diff] [blame] | 148 | if (position > -1) { |
| 149 | for (i = 15; i > position; i--) { |
bjornv@webrtc.org | eec739f | 2012-06-11 07:57:57 +0000 | [diff] [blame] | 150 | smallest_values[i] = smallest_values[i - 1]; |
| 151 | age[i] = age[i - 1]; |
niklase@google.com | 470e71d | 2011-07-07 08:21:25 +0000 | [diff] [blame] | 152 | } |
bjornv@webrtc.org | eec739f | 2012-06-11 07:57:57 +0000 | [diff] [blame] | 153 | smallest_values[position] = feature_value; |
| 154 | age[position] = 1; |
bjornv@webrtc.org | 226c5a1 | 2012-01-04 09:15:12 +0000 | [diff] [blame] | 155 | } |
niklase@google.com | 470e71d | 2011-07-07 08:21:25 +0000 | [diff] [blame] | 156 | |
bjornv@webrtc.org | 226c5a1 | 2012-01-04 09:15:12 +0000 | [diff] [blame] | 157 | // Get |current_median|. |
| 158 | if (self->frame_counter > 2) { |
bjornv@webrtc.org | eec739f | 2012-06-11 07:57:57 +0000 | [diff] [blame] | 159 | current_median = smallest_values[2]; |
bjornv@webrtc.org | 226c5a1 | 2012-01-04 09:15:12 +0000 | [diff] [blame] | 160 | } else if (self->frame_counter > 0) { |
bjornv@webrtc.org | eec739f | 2012-06-11 07:57:57 +0000 | [diff] [blame] | 161 | current_median = smallest_values[0]; |
bjornv@webrtc.org | 226c5a1 | 2012-01-04 09:15:12 +0000 | [diff] [blame] | 162 | } |
| 163 | |
| 164 | // Smooth the median value. |
| 165 | if (self->frame_counter > 0) { |
| 166 | if (current_median < self->mean_value[channel]) { |
bjornv@webrtc.org | a496b03 | 2012-03-20 12:53:06 +0000 | [diff] [blame] | 167 | alpha = kSmoothingDown; // 0.2 in Q15. |
bjornv@webrtc.org | 226c5a1 | 2012-01-04 09:15:12 +0000 | [diff] [blame] | 168 | } else { |
bjornv@webrtc.org | a496b03 | 2012-03-20 12:53:06 +0000 | [diff] [blame] | 169 | alpha = kSmoothingUp; // 0.99 in Q15. |
niklase@google.com | 470e71d | 2011-07-07 08:21:25 +0000 | [diff] [blame] | 170 | } |
bjornv@webrtc.org | 226c5a1 | 2012-01-04 09:15:12 +0000 | [diff] [blame] | 171 | } |
bjornv@webrtc.org | d25c034 | 2015-01-26 15:32:47 +0000 | [diff] [blame] | 172 | tmp32 = (alpha + 1) * self->mean_value[channel]; |
| 173 | tmp32 += (WEBRTC_SPL_WORD16_MAX - alpha) * current_median; |
bjornv@webrtc.org | 226c5a1 | 2012-01-04 09:15:12 +0000 | [diff] [blame] | 174 | tmp32 += 16384; |
| 175 | self->mean_value[channel] = (int16_t) (tmp32 >> 15); |
niklase@google.com | 470e71d | 2011-07-07 08:21:25 +0000 | [diff] [blame] | 176 | |
bjornv@webrtc.org | 226c5a1 | 2012-01-04 09:15:12 +0000 | [diff] [blame] | 177 | return self->mean_value[channel]; |
niklase@google.com | 470e71d | 2011-07-07 08:21:25 +0000 | [diff] [blame] | 178 | } |