1 /* replaygain_synthesis - Routines for applying ReplayGain to a signal
2 * Copyright (C) 2002-2009 Josh Coalson
3 * Copyright (C) 2011-2016 Xiph.Org Foundation
4 *
5 * This library is free software; you can redistribute it and/or
6 * modify it under the terms of the GNU Lesser General Public
7 * License as published by the Free Software Foundation; either
8 * version 2.1 of the License, or (at your option) any later version.
9 *
10 * This library is distributed in the hope that it will be useful,
11 * but WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
13 * Lesser General Public License for more details.
14 *
15 * You should have received a copy of the GNU Lesser General Public
16 * License along with this library; if not, write to the Free Software
17 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
18 */
19 /*
20 * This is an aggregation of pieces of code from John Edwards' WaveGain
21 * program. Mostly cosmetic changes were made; otherwise, the dithering
22 * code is almost untouched and the gain processing was converted from
23 * processing a whole file to processing chunks of samples.
24 *
25 * The original copyright notices for WaveGain's dither.c and wavegain.c
26 * appear below:
27 */
28 /*
29 * (c) 2002 John Edwards
30 * mostly lifted from work by Frank Klemm
31 * random functions for dithering.
32 */
33 /*
34 * Copyright (C) 2002 John Edwards
35 * Additional code by Magnus Holmgren and Gian-Carlo Pascutto
36 */
37
38 #ifdef HAVE_CONFIG_H
39 # include <config.h>
40 #endif
41
42 #include <string.h> /* for memset() */
43 #include <math.h>
44 #include "share/replaygain_synthesis.h"
45 #include "FLAC/assert.h"
46
47 #define FLAC__I64L(x) x##LL
48
49
50 /*
51 * the following is based on parts of dither.c
52 */
53
54
55 /*
56 * This is a simple random number generator with good quality for audio purposes.
57 * It consists of two polycounters with opposite rotation direction and different
58 * periods. The periods are coprime, so the total period is the product of both.
59 *
60 * -------------------------------------------------------------------------------------------------
61 * +-> |31:30:29:28:27:26:25:24:23:22:21:20:19:18:17:16:15:14:13:12:11:10: 9: 8: 7: 6: 5: 4: 3: 2: 1: 0|
62 * | -------------------------------------------------------------------------------------------------
63 * | | | | | | |
64 * | +--+--+--+-XOR-+--------+
65 * | |
66 * +--------------------------------------------------------------------------------------+
67 *
68 * -------------------------------------------------------------------------------------------------
69 * |31:30:29:28:27:26:25:24:23:22:21:20:19:18:17:16:15:14:13:12:11:10: 9: 8: 7: 6: 5: 4: 3: 2: 1: 0| <-+
70 * ------------------------------------------------------------------------------------------------- |
71 * | | | | |
72 * +--+----XOR----+--+ |
73 * | |
74 * +----------------------------------------------------------------------------------------+
75 *
76 *
77 * The first has an period of 3*5*17*257*65537, the second of 7*47*73*178481,
78 * which gives a period of 18.410.713.077.675.721.215. The result is the
79 * XORed values of both generators.
80 */
81
random_int_(void)82 static unsigned int random_int_(void)
83 {
84 static const unsigned char parity_[256] = {
85 0,1,1,0,1,0,0,1,1,0,0,1,0,1,1,0,1,0,0,1,0,1,1,0,0,1,1,0,1,0,0,1,
86 1,0,0,1,0,1,1,0,0,1,1,0,1,0,0,1,0,1,1,0,1,0,0,1,1,0,0,1,0,1,1,0,
87 1,0,0,1,0,1,1,0,0,1,1,0,1,0,0,1,0,1,1,0,1,0,0,1,1,0,0,1,0,1,1,0,
88 0,1,1,0,1,0,0,1,1,0,0,1,0,1,1,0,1,0,0,1,0,1,1,0,0,1,1,0,1,0,0,1,
89 1,0,0,1,0,1,1,0,0,1,1,0,1,0,0,1,0,1,1,0,1,0,0,1,1,0,0,1,0,1,1,0,
90 0,1,1,0,1,0,0,1,1,0,0,1,0,1,1,0,1,0,0,1,0,1,1,0,0,1,1,0,1,0,0,1,
91 0,1,1,0,1,0,0,1,1,0,0,1,0,1,1,0,1,0,0,1,0,1,1,0,0,1,1,0,1,0,0,1,
92 1,0,0,1,0,1,1,0,0,1,1,0,1,0,0,1,0,1,1,0,1,0,0,1,1,0,0,1,0,1,1,0
93 };
94 static unsigned int r1_ = 1;
95 static unsigned int r2_ = 1;
96
97 unsigned int t1, t2, t3, t4;
98
99 /* Parity calculation is done via table lookup, this is also available
100 * on CPUs without parity, can be implemented in C and avoid unpredictable
101 * jumps and slow rotate through the carry flag operations.
102 */
103 t3 = t1 = r1_; t4 = t2 = r2_;
104 t1 &= 0xF5; t2 >>= 25;
105 t1 = parity_[t1]; t2 &= 0x63;
106 t1 <<= 31; t2 = parity_[t2];
107
108 return (r1_ = (t3 >> 1) | t1 ) ^ (r2_ = (t4 + t4) | t2 );
109 }
110
111 /* gives a equal distributed random number */
112 /* between -2^31*mult and +2^31*mult */
random_equi_(double mult)113 static double random_equi_(double mult)
114 {
115 return mult * (int) random_int_();
116 }
117
118 /* gives a triangular distributed random number */
119 /* between -2^32*mult and +2^32*mult */
random_triangular_(double mult)120 static double random_triangular_(double mult)
121 {
122 return mult * ( (double) (int) random_int_() + (double) (int) random_int_() );
123 }
124
125
126 static const float F44_0 [16 + 32] = {
127 (float)0, (float)0, (float)0, (float)0, (float)0, (float)0, (float)0, (float)0,
128 (float)0, (float)0, (float)0, (float)0, (float)0, (float)0, (float)0, (float)0,
129
130 (float)0, (float)0, (float)0, (float)0, (float)0, (float)0, (float)0, (float)0,
131 (float)0, (float)0, (float)0, (float)0, (float)0, (float)0, (float)0, (float)0,
132
133 (float)0, (float)0, (float)0, (float)0, (float)0, (float)0, (float)0, (float)0,
134 (float)0, (float)0, (float)0, (float)0, (float)0, (float)0, (float)0, (float)0
135 };
136
137
138 static const float F44_1 [16 + 32] = { /* SNR(w) = 4.843163 dB, SNR = -3.192134 dB */
139 (float) 0.85018292704024355931, (float) 0.29089597350995344721, (float)-0.05021866022121039450, (float)-0.23545456294599161833,
140 (float)-0.58362726442227032096, (float)-0.67038978965193036429, (float)-0.38566861572833459221, (float)-0.15218663390367969967,
141 (float)-0.02577543084864530676, (float) 0.14119295297688728127, (float) 0.22398848581628781612, (float) 0.15401727203382084116,
142 (float) 0.05216161232906000929, (float)-0.00282237820999675451, (float)-0.03042794608323867363, (float)-0.03109780942998826024,
143
144 (float) 0.85018292704024355931, (float) 0.29089597350995344721, (float)-0.05021866022121039450, (float)-0.23545456294599161833,
145 (float)-0.58362726442227032096, (float)-0.67038978965193036429, (float)-0.38566861572833459221, (float)-0.15218663390367969967,
146 (float)-0.02577543084864530676, (float) 0.14119295297688728127, (float) 0.22398848581628781612, (float) 0.15401727203382084116,
147 (float) 0.05216161232906000929, (float)-0.00282237820999675451, (float)-0.03042794608323867363, (float)-0.03109780942998826024,
148
149 (float) 0.85018292704024355931, (float) 0.29089597350995344721, (float)-0.05021866022121039450, (float)-0.23545456294599161833,
150 (float)-0.58362726442227032096, (float)-0.67038978965193036429, (float)-0.38566861572833459221, (float)-0.15218663390367969967,
151 (float)-0.02577543084864530676, (float) 0.14119295297688728127, (float) 0.22398848581628781612, (float) 0.15401727203382084116,
152 (float) 0.05216161232906000929, (float)-0.00282237820999675451, (float)-0.03042794608323867363, (float)-0.03109780942998826024,
153 };
154
155
156 static const float F44_2 [16 + 32] = { /* SNR(w) = 10.060213 dB, SNR = -12.766730 dB */
157 (float) 1.78827593892108555290, (float) 0.95508210637394326553, (float)-0.18447626783899924429, (float)-0.44198126506275016437,
158 (float)-0.88404052492547413497, (float)-1.42218907262407452967, (float)-1.02037566838362314995, (float)-0.34861755756425577264,
159 (float)-0.11490230170431934434, (float) 0.12498899339968611803, (float) 0.38065885268563131927, (float) 0.31883491321310506562,
160 (float) 0.10486838686563442765, (float)-0.03105361685110374845, (float)-0.06450524884075370758, (float)-0.02939198261121969816,
161
162 (float) 1.78827593892108555290, (float) 0.95508210637394326553, (float)-0.18447626783899924429, (float)-0.44198126506275016437,
163 (float)-0.88404052492547413497, (float)-1.42218907262407452967, (float)-1.02037566838362314995, (float)-0.34861755756425577264,
164 (float)-0.11490230170431934434, (float) 0.12498899339968611803, (float) 0.38065885268563131927, (float) 0.31883491321310506562,
165 (float) 0.10486838686563442765, (float)-0.03105361685110374845, (float)-0.06450524884075370758, (float)-0.02939198261121969816,
166
167 (float) 1.78827593892108555290, (float) 0.95508210637394326553, (float)-0.18447626783899924429, (float)-0.44198126506275016437,
168 (float)-0.88404052492547413497, (float)-1.42218907262407452967, (float)-1.02037566838362314995, (float)-0.34861755756425577264,
169 (float)-0.11490230170431934434, (float) 0.12498899339968611803, (float) 0.38065885268563131927, (float) 0.31883491321310506562,
170 (float) 0.10486838686563442765, (float)-0.03105361685110374845, (float)-0.06450524884075370758, (float)-0.02939198261121969816,
171 };
172
173
174 static const float F44_3 [16 + 32] = { /* SNR(w) = 15.382598 dB, SNR = -29.402334 dB */
175 (float) 2.89072132015058161445, (float) 2.68932810943698754106, (float) 0.21083359339410251227, (float)-0.98385073324997617515,
176 (float)-1.11047823227097316719, (float)-2.18954076314139673147, (float)-2.36498032881953056225, (float)-0.95484132880101140785,
177 (float)-0.23924057925542965158, (float)-0.13865235703915925642, (float) 0.43587843191057992846, (float) 0.65903257226026665927,
178 (float) 0.24361815372443152787, (float)-0.00235974960154720097, (float) 0.01844166574603346289, (float) 0.01722945988740875099,
179
180 (float) 2.89072132015058161445, (float) 2.68932810943698754106, (float) 0.21083359339410251227, (float)-0.98385073324997617515,
181 (float)-1.11047823227097316719, (float)-2.18954076314139673147, (float)-2.36498032881953056225, (float)-0.95484132880101140785,
182 (float)-0.23924057925542965158, (float)-0.13865235703915925642, (float) 0.43587843191057992846, (float) 0.65903257226026665927,
183 (float) 0.24361815372443152787, (float)-0.00235974960154720097, (float) 0.01844166574603346289, (float) 0.01722945988740875099,
184
185 (float) 2.89072132015058161445, (float) 2.68932810943698754106, (float) 0.21083359339410251227, (float)-0.98385073324997617515,
186 (float)-1.11047823227097316719, (float)-2.18954076314139673147, (float)-2.36498032881953056225, (float)-0.95484132880101140785,
187 (float)-0.23924057925542965158, (float)-0.13865235703915925642, (float) 0.43587843191057992846, (float) 0.65903257226026665927,
188 (float) 0.24361815372443152787, (float)-0.00235974960154720097, (float) 0.01844166574603346289, (float) 0.01722945988740875099
189 };
190
191
scalar16_(const float * x,const float * y)192 static double scalar16_(const float* x, const float* y)
193 {
194 return
195 x[ 0]*y[ 0] + x[ 1]*y[ 1] + x[ 2]*y[ 2] + x[ 3]*y[ 3] +
196 x[ 4]*y[ 4] + x[ 5]*y[ 5] + x[ 6]*y[ 6] + x[ 7]*y[ 7] +
197 x[ 8]*y[ 8] + x[ 9]*y[ 9] + x[10]*y[10] + x[11]*y[11] +
198 x[12]*y[12] + x[13]*y[13] + x[14]*y[14] + x[15]*y[15];
199 }
200
201
FLAC__replaygain_synthesis__init_dither_context(DitherContext * d,int bits,int shapingtype)202 void FLAC__replaygain_synthesis__init_dither_context(DitherContext *d, int bits, int shapingtype)
203 {
204 static unsigned char default_dither [] = { 92, 92, 88, 84, 81, 78, 74, 67, 0, 0 };
205 static const float* F [] = { F44_0, F44_1, F44_2, F44_3 };
206
207 int indx;
208
209 if (shapingtype < 0) shapingtype = 0;
210 if (shapingtype > 3) shapingtype = 3;
211 d->ShapingType = (NoiseShaping)shapingtype;
212 indx = bits - 11 - shapingtype;
213 if (indx < 0) indx = 0;
214 if (indx > 9) indx = 9;
215
216 memset ( d->ErrorHistory , 0, sizeof (d->ErrorHistory ) );
217 memset ( d->DitherHistory, 0, sizeof (d->DitherHistory) );
218
219 d->FilterCoeff = F [shapingtype];
220 d->Mask = ((FLAC__uint64)-1) << (32 - bits);
221 d->Add = 0.5 * ((1L << (32 - bits)) - 1);
222 d->Dither = 0.01f*default_dither[indx] / (((FLAC__int64)1) << bits);
223 d->LastHistoryIndex = 0;
224 }
225
226 /*
227 * the following is based on parts of wavegain.c
228 */
229
dither_output_(DitherContext * d,FLAC__bool do_dithering,int shapingtype,int i,double Sum,int k)230 static FLAC__int64 dither_output_(DitherContext *d, FLAC__bool do_dithering, int shapingtype, int i, double Sum, int k)
231 {
232 union {
233 double d;
234 FLAC__int64 i;
235 } doubletmp;
236 double Sum2;
237 FLAC__int64 val;
238
239 #define ROUND64(x) ( doubletmp.d = (x) + d->Add + (FLAC__int64)FLAC__I64L(0x001FFFFD80000000), doubletmp.i - (FLAC__int64)FLAC__I64L(0x433FFFFD80000000) )
240
241 if(do_dithering) {
242 if(shapingtype == 0) {
243 double tmp = random_equi_(d->Dither);
244 Sum2 = tmp - d->LastRandomNumber [k];
245 d->LastRandomNumber [k] = (int)tmp;
246 Sum2 = Sum += Sum2;
247 val = ROUND64(Sum2) & d->Mask;
248 }
249 else {
250 Sum2 = random_triangular_(d->Dither) - scalar16_(d->DitherHistory[k], d->FilterCoeff + i);
251 Sum += d->DitherHistory [k] [(-1-i)&15] = (float)Sum2;
252 Sum2 = Sum + scalar16_(d->ErrorHistory [k], d->FilterCoeff + i);
253 val = ROUND64(Sum2) & d->Mask;
254 d->ErrorHistory [k] [(-1-i)&15] = (float)(Sum - val);
255 }
256 return val;
257 }
258 else
259 return ROUND64(Sum);
260
261 #undef ROUND64
262 }
263
264 #if 0
265 float peak = 0.f,
266 new_peak,
267 factor_clip
268 double scale,
269 dB;
270
271 ...
272
273 peak is in the range -32768.0 .. 32767.0
274
275 /* calculate factors for ReplayGain and ClippingPrevention */
276 *track_gain = GetTitleGain() + settings->man_gain;
277 scale = (float) pow(10., *track_gain * 0.05);
278 if(settings->clip_prev) {
279 factor_clip = (float) (32767./( peak + 1));
280 if(scale < factor_clip)
281 factor_clip = 1.f;
282 else
283 factor_clip /= scale;
284 scale *= factor_clip;
285 }
286 new_peak = (float) peak * scale;
287
288 dB = 20. * log10(scale);
289 *track_gain = (float) dB;
290
291 const double scale = pow(10., (double)gain * 0.05);
292 #endif
293
294
FLAC__replaygain_synthesis__apply_gain(FLAC__byte * data_out,FLAC__bool little_endian_data_out,FLAC__bool unsigned_data_out,const FLAC__int32 * const input[],unsigned wide_samples,unsigned channels,const unsigned source_bps,const unsigned target_bps,const double scale,const FLAC__bool hard_limit,FLAC__bool do_dithering,DitherContext * dither_context)295 size_t FLAC__replaygain_synthesis__apply_gain(FLAC__byte *data_out, FLAC__bool little_endian_data_out, FLAC__bool unsigned_data_out, const FLAC__int32 * const input[], unsigned wide_samples, unsigned channels, const unsigned source_bps, const unsigned target_bps, const double scale, const FLAC__bool hard_limit, FLAC__bool do_dithering, DitherContext *dither_context)
296 {
297 static const FLAC__int64 hard_clip_factors_[33] = {
298 0, /* 0 bits-per-sample (not supported) */
299 0, /* 1 bits-per-sample (not supported) */
300 0, /* 2 bits-per-sample (not supported) */
301 0, /* 3 bits-per-sample (not supported) */
302 -8, /* 4 bits-per-sample */
303 -16, /* 5 bits-per-sample */
304 -32, /* 6 bits-per-sample */
305 -64, /* 7 bits-per-sample */
306 -128, /* 8 bits-per-sample */
307 -256, /* 9 bits-per-sample */
308 -512, /* 10 bits-per-sample */
309 -1024, /* 11 bits-per-sample */
310 -2048, /* 12 bits-per-sample */
311 -4096, /* 13 bits-per-sample */
312 -8192, /* 14 bits-per-sample */
313 -16384, /* 15 bits-per-sample */
314 -32768, /* 16 bits-per-sample */
315 -65536, /* 17 bits-per-sample */
316 -131072, /* 18 bits-per-sample */
317 -262144, /* 19 bits-per-sample */
318 -524288, /* 20 bits-per-sample */
319 -1048576, /* 21 bits-per-sample */
320 -2097152, /* 22 bits-per-sample */
321 -4194304, /* 23 bits-per-sample */
322 -8388608, /* 24 bits-per-sample */
323 -16777216, /* 25 bits-per-sample */
324 -33554432, /* 26 bits-per-sample */
325 -67108864, /* 27 bits-per-sample */
326 -134217728, /* 28 bits-per-sample */
327 -268435456, /* 29 bits-per-sample */
328 -536870912, /* 30 bits-per-sample */
329 -1073741824, /* 31 bits-per-sample */
330 (FLAC__int64)(-1073741824) * 2 /* 32 bits-per-sample */
331 };
332 const FLAC__int32 conv_shift = 32 - target_bps;
333 const FLAC__int64 hard_clip_factor = hard_clip_factors_[target_bps];
334 /*
335 * The integer input coming in has a varying range based on the
336 * source_bps. We want to normalize it to [-1.0, 1.0) so instead
337 * of doing two multiplies on each sample, we just multiple
338 * 'scale' by 1/(2^(source_bps-1))
339 */
340 const double multi_scale = scale / (double)(1u << (source_bps-1));
341
342 FLAC__byte * const start = data_out;
343 unsigned i, channel;
344 const FLAC__int32 *input_;
345 double sample;
346 const unsigned bytes_per_sample = target_bps / 8;
347 const unsigned last_history_index = dither_context->LastHistoryIndex;
348 NoiseShaping noise_shaping = dither_context->ShapingType;
349 FLAC__int64 val64;
350 FLAC__int32 val32;
351 FLAC__int32 uval32;
352 const FLAC__uint32 twiggle = 1u << (target_bps - 1);
353
354 FLAC__ASSERT(channels > 0 && channels <= FLAC_SHARE__MAX_SUPPORTED_CHANNELS);
355 FLAC__ASSERT(source_bps >= 4);
356 FLAC__ASSERT(target_bps >= 4);
357 FLAC__ASSERT(source_bps <= 32);
358 FLAC__ASSERT(target_bps < 32);
359 FLAC__ASSERT((target_bps & 7) == 0);
360
361 for(channel = 0; channel < channels; channel++) {
362 const unsigned incr = bytes_per_sample * channels;
363 data_out = start + bytes_per_sample * channel;
364 input_ = input[channel];
365 for(i = 0; i < wide_samples; i++, data_out += incr) {
366 sample = (double)input_[i] * multi_scale;
367
368 if(hard_limit) {
369 /* hard 6dB limiting */
370 if(sample < -0.5)
371 sample = tanh((sample + 0.5) / (1-0.5)) * (1-0.5) - 0.5;
372 else if(sample > 0.5)
373 sample = tanh((sample - 0.5) / (1-0.5)) * (1-0.5) + 0.5;
374 }
375 sample *= 2147483647.;
376
377 val64 = dither_output_(dither_context, do_dithering, noise_shaping, (i + last_history_index) % 32, sample, channel) >> conv_shift;
378
379 val32 = (FLAC__int32)val64;
380 if(val64 >= -hard_clip_factor)
381 val32 = (FLAC__int32)(-(hard_clip_factor+1));
382 else if(val64 < hard_clip_factor)
383 val32 = (FLAC__int32)hard_clip_factor;
384
385 uval32 = (FLAC__uint32)val32;
386 if (unsigned_data_out)
387 uval32 ^= twiggle;
388
389 if (little_endian_data_out) {
390 switch(target_bps) {
391 case 24:
392 data_out[2] = (FLAC__byte)(uval32 >> 16);
393 /* fall through */
394 case 16:
395 data_out[1] = (FLAC__byte)(uval32 >> 8);
396 /* fall through */
397 case 8:
398 data_out[0] = (FLAC__byte)uval32;
399 break;
400 }
401 }
402 else {
403 switch(target_bps) {
404 case 24:
405 data_out[0] = (FLAC__byte)(uval32 >> 16);
406 data_out[1] = (FLAC__byte)(uval32 >> 8);
407 data_out[2] = (FLAC__byte)uval32;
408 break;
409 case 16:
410 data_out[0] = (FLAC__byte)(uval32 >> 8);
411 data_out[1] = (FLAC__byte)uval32;
412 break;
413 case 8:
414 data_out[0] = (FLAC__byte)uval32;
415 break;
416 }
417 }
418 }
419 }
420 dither_context->LastHistoryIndex = (last_history_index + wide_samples) % 32;
421
422 return wide_samples * channels * (target_bps/8);
423 }
424