media/libopus/silk/stereo_LR_to_MS.c

Thu, 22 Jan 2015 13:21:57 +0100

author
Michael Schloh von Bennewitz <michael@schloh.com>
date
Thu, 22 Jan 2015 13:21:57 +0100
branch
TOR_BUG_9701
changeset 15
b8a032363ba2
permissions
-rw-r--r--

Incorporate requested changes from Mozilla in review:
https://bugzilla.mozilla.org/show_bug.cgi?id=1123480#c6

michael@0 1 /***********************************************************************
michael@0 2 Copyright (c) 2006-2011, Skype Limited. All rights reserved.
michael@0 3 Redistribution and use in source and binary forms, with or without
michael@0 4 modification, are permitted provided that the following conditions
michael@0 5 are met:
michael@0 6 - Redistributions of source code must retain the above copyright notice,
michael@0 7 this list of conditions and the following disclaimer.
michael@0 8 - Redistributions in binary form must reproduce the above copyright
michael@0 9 notice, this list of conditions and the following disclaimer in the
michael@0 10 documentation and/or other materials provided with the distribution.
michael@0 11 - Neither the name of Internet Society, IETF or IETF Trust, nor the
michael@0 12 names of specific contributors, may be used to endorse or promote
michael@0 13 products derived from this software without specific prior written
michael@0 14 permission.
michael@0 15 THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
michael@0 16 AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
michael@0 17 IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
michael@0 18 ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE
michael@0 19 LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
michael@0 20 CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
michael@0 21 SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
michael@0 22 INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
michael@0 23 CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
michael@0 24 ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
michael@0 25 POSSIBILITY OF SUCH DAMAGE.
michael@0 26 ***********************************************************************/
michael@0 27
michael@0 28 #ifdef HAVE_CONFIG_H
michael@0 29 #include "config.h"
michael@0 30 #endif
michael@0 31
michael@0 32 #include "main.h"
michael@0 33 #include "stack_alloc.h"
michael@0 34
michael@0 35 /* Convert Left/Right stereo signal to adaptive Mid/Side representation */
michael@0 36 void silk_stereo_LR_to_MS(
michael@0 37 stereo_enc_state *state, /* I/O State */
michael@0 38 opus_int16 x1[], /* I/O Left input signal, becomes mid signal */
michael@0 39 opus_int16 x2[], /* I/O Right input signal, becomes side signal */
michael@0 40 opus_int8 ix[ 2 ][ 3 ], /* O Quantization indices */
michael@0 41 opus_int8 *mid_only_flag, /* O Flag: only mid signal coded */
michael@0 42 opus_int32 mid_side_rates_bps[], /* O Bitrates for mid and side signals */
michael@0 43 opus_int32 total_rate_bps, /* I Total bitrate */
michael@0 44 opus_int prev_speech_act_Q8, /* I Speech activity level in previous frame */
michael@0 45 opus_int toMono, /* I Last frame before a stereo->mono transition */
michael@0 46 opus_int fs_kHz, /* I Sample rate (kHz) */
michael@0 47 opus_int frame_length /* I Number of samples */
michael@0 48 )
michael@0 49 {
michael@0 50 opus_int n, is10msFrame, denom_Q16, delta0_Q13, delta1_Q13;
michael@0 51 opus_int32 sum, diff, smooth_coef_Q16, pred_Q13[ 2 ], pred0_Q13, pred1_Q13;
michael@0 52 opus_int32 LP_ratio_Q14, HP_ratio_Q14, frac_Q16, frac_3_Q16, min_mid_rate_bps, width_Q14, w_Q24, deltaw_Q24;
michael@0 53 VARDECL( opus_int16, side );
michael@0 54 VARDECL( opus_int16, LP_mid );
michael@0 55 VARDECL( opus_int16, HP_mid );
michael@0 56 VARDECL( opus_int16, LP_side );
michael@0 57 VARDECL( opus_int16, HP_side );
michael@0 58 opus_int16 *mid = &x1[ -2 ];
michael@0 59 SAVE_STACK;
michael@0 60
michael@0 61 ALLOC( side, frame_length + 2, opus_int16 );
michael@0 62 /* Convert to basic mid/side signals */
michael@0 63 for( n = 0; n < frame_length + 2; n++ ) {
michael@0 64 sum = x1[ n - 2 ] + (opus_int32)x2[ n - 2 ];
michael@0 65 diff = x1[ n - 2 ] - (opus_int32)x2[ n - 2 ];
michael@0 66 mid[ n ] = (opus_int16)silk_RSHIFT_ROUND( sum, 1 );
michael@0 67 side[ n ] = (opus_int16)silk_SAT16( silk_RSHIFT_ROUND( diff, 1 ) );
michael@0 68 }
michael@0 69
michael@0 70 /* Buffering */
michael@0 71 silk_memcpy( mid, state->sMid, 2 * sizeof( opus_int16 ) );
michael@0 72 silk_memcpy( side, state->sSide, 2 * sizeof( opus_int16 ) );
michael@0 73 silk_memcpy( state->sMid, &mid[ frame_length ], 2 * sizeof( opus_int16 ) );
michael@0 74 silk_memcpy( state->sSide, &side[ frame_length ], 2 * sizeof( opus_int16 ) );
michael@0 75
michael@0 76 /* LP and HP filter mid signal */
michael@0 77 ALLOC( LP_mid, frame_length, opus_int16 );
michael@0 78 ALLOC( HP_mid, frame_length, opus_int16 );
michael@0 79 for( n = 0; n < frame_length; n++ ) {
michael@0 80 sum = silk_RSHIFT_ROUND( silk_ADD_LSHIFT( mid[ n ] + mid[ n + 2 ], mid[ n + 1 ], 1 ), 2 );
michael@0 81 LP_mid[ n ] = sum;
michael@0 82 HP_mid[ n ] = mid[ n + 1 ] - sum;
michael@0 83 }
michael@0 84
michael@0 85 /* LP and HP filter side signal */
michael@0 86 ALLOC( LP_side, frame_length, opus_int16 );
michael@0 87 ALLOC( HP_side, frame_length, opus_int16 );
michael@0 88 for( n = 0; n < frame_length; n++ ) {
michael@0 89 sum = silk_RSHIFT_ROUND( silk_ADD_LSHIFT( side[ n ] + side[ n + 2 ], side[ n + 1 ], 1 ), 2 );
michael@0 90 LP_side[ n ] = sum;
michael@0 91 HP_side[ n ] = side[ n + 1 ] - sum;
michael@0 92 }
michael@0 93
michael@0 94 /* Find energies and predictors */
michael@0 95 is10msFrame = frame_length == 10 * fs_kHz;
michael@0 96 smooth_coef_Q16 = is10msFrame ?
michael@0 97 SILK_FIX_CONST( STEREO_RATIO_SMOOTH_COEF / 2, 16 ) :
michael@0 98 SILK_FIX_CONST( STEREO_RATIO_SMOOTH_COEF, 16 );
michael@0 99 smooth_coef_Q16 = silk_SMULWB( silk_SMULBB( prev_speech_act_Q8, prev_speech_act_Q8 ), smooth_coef_Q16 );
michael@0 100
michael@0 101 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 );
michael@0 102 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 );
michael@0 103 /* Ratio of the norms of residual and mid signals */
michael@0 104 frac_Q16 = silk_SMLABB( HP_ratio_Q14, LP_ratio_Q14, 3 );
michael@0 105 frac_Q16 = silk_min( frac_Q16, SILK_FIX_CONST( 1, 16 ) );
michael@0 106
michael@0 107 /* Determine bitrate distribution between mid and side, and possibly reduce stereo width */
michael@0 108 total_rate_bps -= is10msFrame ? 1200 : 600; /* Subtract approximate bitrate for coding stereo parameters */
michael@0 109 if( total_rate_bps < 1 ) {
michael@0 110 total_rate_bps = 1;
michael@0 111 }
michael@0 112 min_mid_rate_bps = silk_SMLABB( 2000, fs_kHz, 900 );
michael@0 113 silk_assert( min_mid_rate_bps < 32767 );
michael@0 114 /* Default bitrate distribution: 8 parts for Mid and (5+3*frac) parts for Side. so: mid_rate = ( 8 / ( 13 + 3 * frac ) ) * total_ rate */
michael@0 115 frac_3_Q16 = silk_MUL( 3, frac_Q16 );
michael@0 116 mid_side_rates_bps[ 0 ] = silk_DIV32_varQ( total_rate_bps, SILK_FIX_CONST( 8 + 5, 16 ) + frac_3_Q16, 16+3 );
michael@0 117 /* If Mid bitrate below minimum, reduce stereo width */
michael@0 118 if( mid_side_rates_bps[ 0 ] < min_mid_rate_bps ) {
michael@0 119 mid_side_rates_bps[ 0 ] = min_mid_rate_bps;
michael@0 120 mid_side_rates_bps[ 1 ] = total_rate_bps - mid_side_rates_bps[ 0 ];
michael@0 121 /* width = 4 * ( 2 * side_rate - min_rate ) / ( ( 1 + 3 * frac ) * min_rate ) */
michael@0 122 width_Q14 = silk_DIV32_varQ( silk_LSHIFT( mid_side_rates_bps[ 1 ], 1 ) - min_mid_rate_bps,
michael@0 123 silk_SMULWB( SILK_FIX_CONST( 1, 16 ) + frac_3_Q16, min_mid_rate_bps ), 14+2 );
michael@0 124 width_Q14 = silk_LIMIT( width_Q14, 0, SILK_FIX_CONST( 1, 14 ) );
michael@0 125 } else {
michael@0 126 mid_side_rates_bps[ 1 ] = total_rate_bps - mid_side_rates_bps[ 0 ];
michael@0 127 width_Q14 = SILK_FIX_CONST( 1, 14 );
michael@0 128 }
michael@0 129
michael@0 130 /* Smoother */
michael@0 131 state->smth_width_Q14 = (opus_int16)silk_SMLAWB( state->smth_width_Q14, width_Q14 - state->smth_width_Q14, smooth_coef_Q16 );
michael@0 132
michael@0 133 /* At very low bitrates or for inputs that are nearly amplitude panned, switch to panned-mono coding */
michael@0 134 *mid_only_flag = 0;
michael@0 135 if( toMono ) {
michael@0 136 /* Last frame before stereo->mono transition; collapse stereo width */
michael@0 137 width_Q14 = 0;
michael@0 138 pred_Q13[ 0 ] = 0;
michael@0 139 pred_Q13[ 1 ] = 0;
michael@0 140 silk_stereo_quant_pred( pred_Q13, ix );
michael@0 141 } else if( state->width_prev_Q14 == 0 &&
michael@0 142 ( 8 * total_rate_bps < 13 * min_mid_rate_bps || silk_SMULWB( frac_Q16, state->smth_width_Q14 ) < SILK_FIX_CONST( 0.05, 14 ) ) )
michael@0 143 {
michael@0 144 /* Code as panned-mono; previous frame already had zero width */
michael@0 145 /* Scale down and quantize predictors */
michael@0 146 pred_Q13[ 0 ] = silk_RSHIFT( silk_SMULBB( state->smth_width_Q14, pred_Q13[ 0 ] ), 14 );
michael@0 147 pred_Q13[ 1 ] = silk_RSHIFT( silk_SMULBB( state->smth_width_Q14, pred_Q13[ 1 ] ), 14 );
michael@0 148 silk_stereo_quant_pred( pred_Q13, ix );
michael@0 149 /* Collapse stereo width */
michael@0 150 width_Q14 = 0;
michael@0 151 pred_Q13[ 0 ] = 0;
michael@0 152 pred_Q13[ 1 ] = 0;
michael@0 153 mid_side_rates_bps[ 0 ] = total_rate_bps;
michael@0 154 mid_side_rates_bps[ 1 ] = 0;
michael@0 155 *mid_only_flag = 1;
michael@0 156 } else if( state->width_prev_Q14 != 0 &&
michael@0 157 ( 8 * total_rate_bps < 11 * min_mid_rate_bps || silk_SMULWB( frac_Q16, state->smth_width_Q14 ) < SILK_FIX_CONST( 0.02, 14 ) ) )
michael@0 158 {
michael@0 159 /* Transition to zero-width stereo */
michael@0 160 /* Scale down and quantize predictors */
michael@0 161 pred_Q13[ 0 ] = silk_RSHIFT( silk_SMULBB( state->smth_width_Q14, pred_Q13[ 0 ] ), 14 );
michael@0 162 pred_Q13[ 1 ] = silk_RSHIFT( silk_SMULBB( state->smth_width_Q14, pred_Q13[ 1 ] ), 14 );
michael@0 163 silk_stereo_quant_pred( pred_Q13, ix );
michael@0 164 /* Collapse stereo width */
michael@0 165 width_Q14 = 0;
michael@0 166 pred_Q13[ 0 ] = 0;
michael@0 167 pred_Q13[ 1 ] = 0;
michael@0 168 } else if( state->smth_width_Q14 > SILK_FIX_CONST( 0.95, 14 ) ) {
michael@0 169 /* Full-width stereo coding */
michael@0 170 silk_stereo_quant_pred( pred_Q13, ix );
michael@0 171 width_Q14 = SILK_FIX_CONST( 1, 14 );
michael@0 172 } else {
michael@0 173 /* Reduced-width stereo coding; scale down and quantize predictors */
michael@0 174 pred_Q13[ 0 ] = silk_RSHIFT( silk_SMULBB( state->smth_width_Q14, pred_Q13[ 0 ] ), 14 );
michael@0 175 pred_Q13[ 1 ] = silk_RSHIFT( silk_SMULBB( state->smth_width_Q14, pred_Q13[ 1 ] ), 14 );
michael@0 176 silk_stereo_quant_pred( pred_Q13, ix );
michael@0 177 width_Q14 = state->smth_width_Q14;
michael@0 178 }
michael@0 179
michael@0 180 /* Make sure to keep on encoding until the tapered output has been transmitted */
michael@0 181 if( *mid_only_flag == 1 ) {
michael@0 182 state->silent_side_len += frame_length - STEREO_INTERP_LEN_MS * fs_kHz;
michael@0 183 if( state->silent_side_len < LA_SHAPE_MS * fs_kHz ) {
michael@0 184 *mid_only_flag = 0;
michael@0 185 } else {
michael@0 186 /* Limit to avoid wrapping around */
michael@0 187 state->silent_side_len = 10000;
michael@0 188 }
michael@0 189 } else {
michael@0 190 state->silent_side_len = 0;
michael@0 191 }
michael@0 192
michael@0 193 if( *mid_only_flag == 0 && mid_side_rates_bps[ 1 ] < 1 ) {
michael@0 194 mid_side_rates_bps[ 1 ] = 1;
michael@0 195 mid_side_rates_bps[ 0 ] = silk_max_int( 1, total_rate_bps - mid_side_rates_bps[ 1 ]);
michael@0 196 }
michael@0 197
michael@0 198 /* Interpolate predictors and subtract prediction from side channel */
michael@0 199 pred0_Q13 = -state->pred_prev_Q13[ 0 ];
michael@0 200 pred1_Q13 = -state->pred_prev_Q13[ 1 ];
michael@0 201 w_Q24 = silk_LSHIFT( state->width_prev_Q14, 10 );
michael@0 202 denom_Q16 = silk_DIV32_16( (opus_int32)1 << 16, STEREO_INTERP_LEN_MS * fs_kHz );
michael@0 203 delta0_Q13 = -silk_RSHIFT_ROUND( silk_SMULBB( pred_Q13[ 0 ] - state->pred_prev_Q13[ 0 ], denom_Q16 ), 16 );
michael@0 204 delta1_Q13 = -silk_RSHIFT_ROUND( silk_SMULBB( pred_Q13[ 1 ] - state->pred_prev_Q13[ 1 ], denom_Q16 ), 16 );
michael@0 205 deltaw_Q24 = silk_LSHIFT( silk_SMULWB( width_Q14 - state->width_prev_Q14, denom_Q16 ), 10 );
michael@0 206 for( n = 0; n < STEREO_INTERP_LEN_MS * fs_kHz; n++ ) {
michael@0 207 pred0_Q13 += delta0_Q13;
michael@0 208 pred1_Q13 += delta1_Q13;
michael@0 209 w_Q24 += deltaw_Q24;
michael@0 210 sum = silk_LSHIFT( silk_ADD_LSHIFT( mid[ n ] + mid[ n + 2 ], mid[ n + 1 ], 1 ), 9 ); /* Q11 */
michael@0 211 sum = silk_SMLAWB( silk_SMULWB( w_Q24, side[ n + 1 ] ), sum, pred0_Q13 ); /* Q8 */
michael@0 212 sum = silk_SMLAWB( sum, silk_LSHIFT( (opus_int32)mid[ n + 1 ], 11 ), pred1_Q13 ); /* Q8 */
michael@0 213 x2[ n - 1 ] = (opus_int16)silk_SAT16( silk_RSHIFT_ROUND( sum, 8 ) );
michael@0 214 }
michael@0 215
michael@0 216 pred0_Q13 = -pred_Q13[ 0 ];
michael@0 217 pred1_Q13 = -pred_Q13[ 1 ];
michael@0 218 w_Q24 = silk_LSHIFT( width_Q14, 10 );
michael@0 219 for( n = STEREO_INTERP_LEN_MS * fs_kHz; n < frame_length; n++ ) {
michael@0 220 sum = silk_LSHIFT( silk_ADD_LSHIFT( mid[ n ] + mid[ n + 2 ], mid[ n + 1 ], 1 ), 9 ); /* Q11 */
michael@0 221 sum = silk_SMLAWB( silk_SMULWB( w_Q24, side[ n + 1 ] ), sum, pred0_Q13 ); /* Q8 */
michael@0 222 sum = silk_SMLAWB( sum, silk_LSHIFT( (opus_int32)mid[ n + 1 ], 11 ), pred1_Q13 ); /* Q8 */
michael@0 223 x2[ n - 1 ] = (opus_int16)silk_SAT16( silk_RSHIFT_ROUND( sum, 8 ) );
michael@0 224 }
michael@0 225 state->pred_prev_Q13[ 0 ] = (opus_int16)pred_Q13[ 0 ];
michael@0 226 state->pred_prev_Q13[ 1 ] = (opus_int16)pred_Q13[ 1 ];
michael@0 227 state->width_prev_Q14 = (opus_int16)width_Q14;
michael@0 228 RESTORE_STACK;
michael@0 229 }

mercurial