annotate src/opus-1.3/silk/stereo_LR_to_MS.c @ 79:91c729825bca pa_catalina

Update build for AUDIO_COMPONENT_FIX
author Chris Cannam
date Wed, 30 Oct 2019 12:40:34 +0000
parents 7aeed7906520
children
rev   line source
Chris@69 1 /***********************************************************************
Chris@69 2 Copyright (c) 2006-2011, Skype Limited. All rights reserved.
Chris@69 3 Redistribution and use in source and binary forms, with or without
Chris@69 4 modification, are permitted provided that the following conditions
Chris@69 5 are met:
Chris@69 6 - Redistributions of source code must retain the above copyright notice,
Chris@69 7 this list of conditions and the following disclaimer.
Chris@69 8 - Redistributions in binary form must reproduce the above copyright
Chris@69 9 notice, this list of conditions and the following disclaimer in the
Chris@69 10 documentation and/or other materials provided with the distribution.
Chris@69 11 - Neither the name of Internet Society, IETF or IETF Trust, nor the
Chris@69 12 names of specific contributors, may be used to endorse or promote
Chris@69 13 products derived from this software without specific prior written
Chris@69 14 permission.
Chris@69 15 THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
Chris@69 16 AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
Chris@69 17 IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
Chris@69 18 ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE
Chris@69 19 LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
Chris@69 20 CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
Chris@69 21 SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
Chris@69 22 INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
Chris@69 23 CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
Chris@69 24 ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
Chris@69 25 POSSIBILITY OF SUCH DAMAGE.
Chris@69 26 ***********************************************************************/
Chris@69 27
Chris@69 28 #ifdef HAVE_CONFIG_H
Chris@69 29 #include "config.h"
Chris@69 30 #endif
Chris@69 31
Chris@69 32 #include "main.h"
Chris@69 33 #include "stack_alloc.h"
Chris@69 34
Chris@69 35 /* Convert Left/Right stereo signal to adaptive Mid/Side representation */
Chris@69 36 void silk_stereo_LR_to_MS(
Chris@69 37 stereo_enc_state *state, /* I/O State */
Chris@69 38 opus_int16 x1[], /* I/O Left input signal, becomes mid signal */
Chris@69 39 opus_int16 x2[], /* I/O Right input signal, becomes side signal */
Chris@69 40 opus_int8 ix[ 2 ][ 3 ], /* O Quantization indices */
Chris@69 41 opus_int8 *mid_only_flag, /* O Flag: only mid signal coded */
Chris@69 42 opus_int32 mid_side_rates_bps[], /* O Bitrates for mid and side signals */
Chris@69 43 opus_int32 total_rate_bps, /* I Total bitrate */
Chris@69 44 opus_int prev_speech_act_Q8, /* I Speech activity level in previous frame */
Chris@69 45 opus_int toMono, /* I Last frame before a stereo->mono transition */
Chris@69 46 opus_int fs_kHz, /* I Sample rate (kHz) */
Chris@69 47 opus_int frame_length /* I Number of samples */
Chris@69 48 )
Chris@69 49 {
Chris@69 50 opus_int n, is10msFrame, denom_Q16, delta0_Q13, delta1_Q13;
Chris@69 51 opus_int32 sum, diff, smooth_coef_Q16, pred_Q13[ 2 ], pred0_Q13, pred1_Q13;
Chris@69 52 opus_int32 LP_ratio_Q14, HP_ratio_Q14, frac_Q16, frac_3_Q16, min_mid_rate_bps, width_Q14, w_Q24, deltaw_Q24;
Chris@69 53 VARDECL( opus_int16, side );
Chris@69 54 VARDECL( opus_int16, LP_mid );
Chris@69 55 VARDECL( opus_int16, HP_mid );
Chris@69 56 VARDECL( opus_int16, LP_side );
Chris@69 57 VARDECL( opus_int16, HP_side );
Chris@69 58 opus_int16 *mid = &x1[ -2 ];
Chris@69 59 SAVE_STACK;
Chris@69 60
Chris@69 61 ALLOC( side, frame_length + 2, opus_int16 );
Chris@69 62 /* Convert to basic mid/side signals */
Chris@69 63 for( n = 0; n < frame_length + 2; n++ ) {
Chris@69 64 sum = x1[ n - 2 ] + (opus_int32)x2[ n - 2 ];
Chris@69 65 diff = x1[ n - 2 ] - (opus_int32)x2[ n - 2 ];
Chris@69 66 mid[ n ] = (opus_int16)silk_RSHIFT_ROUND( sum, 1 );
Chris@69 67 side[ n ] = (opus_int16)silk_SAT16( silk_RSHIFT_ROUND( diff, 1 ) );
Chris@69 68 }
Chris@69 69
Chris@69 70 /* Buffering */
Chris@69 71 silk_memcpy( mid, state->sMid, 2 * sizeof( opus_int16 ) );
Chris@69 72 silk_memcpy( side, state->sSide, 2 * sizeof( opus_int16 ) );
Chris@69 73 silk_memcpy( state->sMid, &mid[ frame_length ], 2 * sizeof( opus_int16 ) );
Chris@69 74 silk_memcpy( state->sSide, &side[ frame_length ], 2 * sizeof( opus_int16 ) );
Chris@69 75
Chris@69 76 /* LP and HP filter mid signal */
Chris@69 77 ALLOC( LP_mid, frame_length, opus_int16 );
Chris@69 78 ALLOC( HP_mid, frame_length, opus_int16 );
Chris@69 79 for( n = 0; n < frame_length; n++ ) {
Chris@69 80 sum = silk_RSHIFT_ROUND( silk_ADD_LSHIFT( mid[ n ] + (opus_int32)mid[ n + 2 ], mid[ n + 1 ], 1 ), 2 );
Chris@69 81 LP_mid[ n ] = sum;
Chris@69 82 HP_mid[ n ] = mid[ n + 1 ] - sum;
Chris@69 83 }
Chris@69 84
Chris@69 85 /* LP and HP filter side signal */
Chris@69 86 ALLOC( LP_side, frame_length, opus_int16 );
Chris@69 87 ALLOC( HP_side, frame_length, opus_int16 );
Chris@69 88 for( n = 0; n < frame_length; n++ ) {
Chris@69 89 sum = silk_RSHIFT_ROUND( silk_ADD_LSHIFT( side[ n ] + (opus_int32)side[ n + 2 ], side[ n + 1 ], 1 ), 2 );
Chris@69 90 LP_side[ n ] = sum;
Chris@69 91 HP_side[ n ] = side[ n + 1 ] - sum;
Chris@69 92 }
Chris@69 93
Chris@69 94 /* Find energies and predictors */
Chris@69 95 is10msFrame = frame_length == 10 * fs_kHz;
Chris@69 96 smooth_coef_Q16 = is10msFrame ?
Chris@69 97 SILK_FIX_CONST( STEREO_RATIO_SMOOTH_COEF / 2, 16 ) :
Chris@69 98 SILK_FIX_CONST( STEREO_RATIO_SMOOTH_COEF, 16 );
Chris@69 99 smooth_coef_Q16 = silk_SMULWB( silk_SMULBB( prev_speech_act_Q8, prev_speech_act_Q8 ), smooth_coef_Q16 );
Chris@69 100
Chris@69 101 pred_Q13[ 0 ] = silk_stereo_find_predictor( &LP_ratio_Q14, LP_mid, LP_side, &state->mid_side_amp_Q0[ 0 ], frame_length, smooth_coef_Q16 );
Chris@69 102 pred_Q13[ 1 ] = silk_stereo_find_predictor( &HP_ratio_Q14, HP_mid, HP_side, &state->mid_side_amp_Q0[ 2 ], frame_length, smooth_coef_Q16 );
Chris@69 103 /* Ratio of the norms of residual and mid signals */
Chris@69 104 frac_Q16 = silk_SMLABB( HP_ratio_Q14, LP_ratio_Q14, 3 );
Chris@69 105 frac_Q16 = silk_min( frac_Q16, SILK_FIX_CONST( 1, 16 ) );
Chris@69 106
Chris@69 107 /* Determine bitrate distribution between mid and side, and possibly reduce stereo width */
Chris@69 108 total_rate_bps -= is10msFrame ? 1200 : 600; /* Subtract approximate bitrate for coding stereo parameters */
Chris@69 109 if( total_rate_bps < 1 ) {
Chris@69 110 total_rate_bps = 1;
Chris@69 111 }
Chris@69 112 min_mid_rate_bps = silk_SMLABB( 2000, fs_kHz, 600 );
Chris@69 113 silk_assert( min_mid_rate_bps < 32767 );
Chris@69 114 /* Default bitrate distribution: 8 parts for Mid and (5+3*frac) parts for Side. so: mid_rate = ( 8 / ( 13 + 3 * frac ) ) * total_ rate */
Chris@69 115 frac_3_Q16 = silk_MUL( 3, frac_Q16 );
Chris@69 116 mid_side_rates_bps[ 0 ] = silk_DIV32_varQ( total_rate_bps, SILK_FIX_CONST( 8 + 5, 16 ) + frac_3_Q16, 16+3 );
Chris@69 117 /* If Mid bitrate below minimum, reduce stereo width */
Chris@69 118 if( mid_side_rates_bps[ 0 ] < min_mid_rate_bps ) {
Chris@69 119 mid_side_rates_bps[ 0 ] = min_mid_rate_bps;
Chris@69 120 mid_side_rates_bps[ 1 ] = total_rate_bps - mid_side_rates_bps[ 0 ];
Chris@69 121 /* width = 4 * ( 2 * side_rate - min_rate ) / ( ( 1 + 3 * frac ) * min_rate ) */
Chris@69 122 width_Q14 = silk_DIV32_varQ( silk_LSHIFT( mid_side_rates_bps[ 1 ], 1 ) - min_mid_rate_bps,
Chris@69 123 silk_SMULWB( SILK_FIX_CONST( 1, 16 ) + frac_3_Q16, min_mid_rate_bps ), 14+2 );
Chris@69 124 width_Q14 = silk_LIMIT( width_Q14, 0, SILK_FIX_CONST( 1, 14 ) );
Chris@69 125 } else {
Chris@69 126 mid_side_rates_bps[ 1 ] = total_rate_bps - mid_side_rates_bps[ 0 ];
Chris@69 127 width_Q14 = SILK_FIX_CONST( 1, 14 );
Chris@69 128 }
Chris@69 129
Chris@69 130 /* Smoother */
Chris@69 131 state->smth_width_Q14 = (opus_int16)silk_SMLAWB( state->smth_width_Q14, width_Q14 - state->smth_width_Q14, smooth_coef_Q16 );
Chris@69 132
Chris@69 133 /* At very low bitrates or for inputs that are nearly amplitude panned, switch to panned-mono coding */
Chris@69 134 *mid_only_flag = 0;
Chris@69 135 if( toMono ) {
Chris@69 136 /* Last frame before stereo->mono transition; collapse stereo width */
Chris@69 137 width_Q14 = 0;
Chris@69 138 pred_Q13[ 0 ] = 0;
Chris@69 139 pred_Q13[ 1 ] = 0;
Chris@69 140 silk_stereo_quant_pred( pred_Q13, ix );
Chris@69 141 } else if( state->width_prev_Q14 == 0 &&
Chris@69 142 ( 8 * total_rate_bps < 13 * min_mid_rate_bps || silk_SMULWB( frac_Q16, state->smth_width_Q14 ) < SILK_FIX_CONST( 0.05, 14 ) ) )
Chris@69 143 {
Chris@69 144 /* Code as panned-mono; previous frame already had zero width */
Chris@69 145 /* Scale down and quantize predictors */
Chris@69 146 pred_Q13[ 0 ] = silk_RSHIFT( silk_SMULBB( state->smth_width_Q14, pred_Q13[ 0 ] ), 14 );
Chris@69 147 pred_Q13[ 1 ] = silk_RSHIFT( silk_SMULBB( state->smth_width_Q14, pred_Q13[ 1 ] ), 14 );
Chris@69 148 silk_stereo_quant_pred( pred_Q13, ix );
Chris@69 149 /* Collapse stereo width */
Chris@69 150 width_Q14 = 0;
Chris@69 151 pred_Q13[ 0 ] = 0;
Chris@69 152 pred_Q13[ 1 ] = 0;
Chris@69 153 mid_side_rates_bps[ 0 ] = total_rate_bps;
Chris@69 154 mid_side_rates_bps[ 1 ] = 0;
Chris@69 155 *mid_only_flag = 1;
Chris@69 156 } else if( state->width_prev_Q14 != 0 &&
Chris@69 157 ( 8 * total_rate_bps < 11 * min_mid_rate_bps || silk_SMULWB( frac_Q16, state->smth_width_Q14 ) < SILK_FIX_CONST( 0.02, 14 ) ) )
Chris@69 158 {
Chris@69 159 /* Transition to zero-width stereo */
Chris@69 160 /* Scale down and quantize predictors */
Chris@69 161 pred_Q13[ 0 ] = silk_RSHIFT( silk_SMULBB( state->smth_width_Q14, pred_Q13[ 0 ] ), 14 );
Chris@69 162 pred_Q13[ 1 ] = silk_RSHIFT( silk_SMULBB( state->smth_width_Q14, pred_Q13[ 1 ] ), 14 );
Chris@69 163 silk_stereo_quant_pred( pred_Q13, ix );
Chris@69 164 /* Collapse stereo width */
Chris@69 165 width_Q14 = 0;
Chris@69 166 pred_Q13[ 0 ] = 0;
Chris@69 167 pred_Q13[ 1 ] = 0;
Chris@69 168 } else if( state->smth_width_Q14 > SILK_FIX_CONST( 0.95, 14 ) ) {
Chris@69 169 /* Full-width stereo coding */
Chris@69 170 silk_stereo_quant_pred( pred_Q13, ix );
Chris@69 171 width_Q14 = SILK_FIX_CONST( 1, 14 );
Chris@69 172 } else {
Chris@69 173 /* Reduced-width stereo coding; scale down and quantize predictors */
Chris@69 174 pred_Q13[ 0 ] = silk_RSHIFT( silk_SMULBB( state->smth_width_Q14, pred_Q13[ 0 ] ), 14 );
Chris@69 175 pred_Q13[ 1 ] = silk_RSHIFT( silk_SMULBB( state->smth_width_Q14, pred_Q13[ 1 ] ), 14 );
Chris@69 176 silk_stereo_quant_pred( pred_Q13, ix );
Chris@69 177 width_Q14 = state->smth_width_Q14;
Chris@69 178 }
Chris@69 179
Chris@69 180 /* Make sure to keep on encoding until the tapered output has been transmitted */
Chris@69 181 if( *mid_only_flag == 1 ) {
Chris@69 182 state->silent_side_len += frame_length - STEREO_INTERP_LEN_MS * fs_kHz;
Chris@69 183 if( state->silent_side_len < LA_SHAPE_MS * fs_kHz ) {
Chris@69 184 *mid_only_flag = 0;
Chris@69 185 } else {
Chris@69 186 /* Limit to avoid wrapping around */
Chris@69 187 state->silent_side_len = 10000;
Chris@69 188 }
Chris@69 189 } else {
Chris@69 190 state->silent_side_len = 0;
Chris@69 191 }
Chris@69 192
Chris@69 193 if( *mid_only_flag == 0 && mid_side_rates_bps[ 1 ] < 1 ) {
Chris@69 194 mid_side_rates_bps[ 1 ] = 1;
Chris@69 195 mid_side_rates_bps[ 0 ] = silk_max_int( 1, total_rate_bps - mid_side_rates_bps[ 1 ]);
Chris@69 196 }
Chris@69 197
Chris@69 198 /* Interpolate predictors and subtract prediction from side channel */
Chris@69 199 pred0_Q13 = -state->pred_prev_Q13[ 0 ];
Chris@69 200 pred1_Q13 = -state->pred_prev_Q13[ 1 ];
Chris@69 201 w_Q24 = silk_LSHIFT( state->width_prev_Q14, 10 );
Chris@69 202 denom_Q16 = silk_DIV32_16( (opus_int32)1 << 16, STEREO_INTERP_LEN_MS * fs_kHz );
Chris@69 203 delta0_Q13 = -silk_RSHIFT_ROUND( silk_SMULBB( pred_Q13[ 0 ] - state->pred_prev_Q13[ 0 ], denom_Q16 ), 16 );
Chris@69 204 delta1_Q13 = -silk_RSHIFT_ROUND( silk_SMULBB( pred_Q13[ 1 ] - state->pred_prev_Q13[ 1 ], denom_Q16 ), 16 );
Chris@69 205 deltaw_Q24 = silk_LSHIFT( silk_SMULWB( width_Q14 - state->width_prev_Q14, denom_Q16 ), 10 );
Chris@69 206 for( n = 0; n < STEREO_INTERP_LEN_MS * fs_kHz; n++ ) {
Chris@69 207 pred0_Q13 += delta0_Q13;
Chris@69 208 pred1_Q13 += delta1_Q13;
Chris@69 209 w_Q24 += deltaw_Q24;
Chris@69 210 sum = silk_LSHIFT( silk_ADD_LSHIFT( mid[ n ] + (opus_int32)mid[ n + 2 ], mid[ n + 1 ], 1 ), 9 ); /* Q11 */
Chris@69 211 sum = silk_SMLAWB( silk_SMULWB( w_Q24, side[ n + 1 ] ), sum, pred0_Q13 ); /* Q8 */
Chris@69 212 sum = silk_SMLAWB( sum, silk_LSHIFT( (opus_int32)mid[ n + 1 ], 11 ), pred1_Q13 ); /* Q8 */
Chris@69 213 x2[ n - 1 ] = (opus_int16)silk_SAT16( silk_RSHIFT_ROUND( sum, 8 ) );
Chris@69 214 }
Chris@69 215
Chris@69 216 pred0_Q13 = -pred_Q13[ 0 ];
Chris@69 217 pred1_Q13 = -pred_Q13[ 1 ];
Chris@69 218 w_Q24 = silk_LSHIFT( width_Q14, 10 );
Chris@69 219 for( n = STEREO_INTERP_LEN_MS * fs_kHz; n < frame_length; n++ ) {
Chris@69 220 sum = silk_LSHIFT( silk_ADD_LSHIFT( mid[ n ] + (opus_int32)mid[ n + 2 ], mid[ n + 1 ], 1 ), 9 ); /* Q11 */
Chris@69 221 sum = silk_SMLAWB( silk_SMULWB( w_Q24, side[ n + 1 ] ), sum, pred0_Q13 ); /* Q8 */
Chris@69 222 sum = silk_SMLAWB( sum, silk_LSHIFT( (opus_int32)mid[ n + 1 ], 11 ), pred1_Q13 ); /* Q8 */
Chris@69 223 x2[ n - 1 ] = (opus_int16)silk_SAT16( silk_RSHIFT_ROUND( sum, 8 ) );
Chris@69 224 }
Chris@69 225 state->pred_prev_Q13[ 0 ] = (opus_int16)pred_Q13[ 0 ];
Chris@69 226 state->pred_prev_Q13[ 1 ] = (opus_int16)pred_Q13[ 1 ];
Chris@69 227 state->width_prev_Q14 = (opus_int16)width_Q14;
Chris@69 228 RESTORE_STACK;
Chris@69 229 }