1 /*
2 * ReplayGainAnalysis - analyzes input samples and give the recommended dB change
3 * Copyright (C) 2001 David Robinson and Glen Sawyer
4 * Improvements and optimizations added by Frank Klemm, and by Marcel Muller
5 *
6 * This library is free software; you can redistribute it and/or
7 * modify it under the terms of the GNU Lesser General Public
8 * License as published by the Free Software Foundation; either
9 * version 2.1 of the License, or (at your option) any later version.
10 *
11 * This library is distributed in the hope that it will be useful,
12 * but WITHOUT ANY WARRANTY; without even the implied warranty of
13 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
14 * Lesser General Public License for more details.
15 *
16 * You should have received a copy of the GNU Lesser General Public
17 * License along with this library; if not, write to the Free Software
18 * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
19 *
20 * concept and filter values by David Robinson (David@Robinson.org)
21 * -- blame him if you think the idea is flawed
22 * original coding by Glen Sawyer (mp3gain@hotmail.com)
23 * -- blame him if you think this runs too slowly, or the coding is otherwise flawed
24 *
25 * lots of code improvements by Frank Klemm ( http://www.uni-jena.de/~pfk/mpp/ )
26 * -- credit him for all the _good_ programming ;)
27 *
28 *
29 * For an explanation of the concepts and the basic algorithms involved, go to:
30 * http://www.replaygain.org/
31 */
32
33 /*
34 * Here's the deal. Call
35 *
36 * InitGainAnalysis ( long samplefreq );
37 *
38 * to initialize everything. Call
39 *
40 * AnalyzeSamples ( const Float_t* left_samples,
41 * const Float_t* right_samples,
42 * size_t num_samples,
43 * int num_channels );
44 *
45 * as many times as you want, with as many or as few samples as you want.
46 * If mono, pass the sample buffer in through left_samples, leave
47 * right_samples NULL, and make sure num_channels = 1.
48 *
49 * GetTitleGain()
50 *
51 * will return the recommended dB level change for all samples analyzed
52 * SINCE THE LAST TIME you called GetTitleGain() OR InitGainAnalysis().
53 *
54 * GetAlbumGain()
55 *
56 * will return the recommended dB level change for all samples analyzed
57 * since InitGainAnalysis() was called and finalized with GetTitleGain().
58 *
59 * Pseudo-code to process an album:
60 *
61 * Float_t l_samples [4096];
62 * Float_t r_samples [4096];
63 * size_t num_samples;
64 * unsigned int num_songs;
65 * unsigned int i;
66 *
67 * InitGainAnalysis ( 44100 );
68 * for ( i = 1; i <= num_songs; i++ ) {
69 * while ( ( num_samples = getSongSamples ( song[i], left_samples, right_samples ) ) > 0 )
70 * AnalyzeSamples ( left_samples, right_samples, num_samples, 2 );
71 * fprintf ("Recommended dB change for song %2d: %+6.2f dB\n", i, GetTitleGain() );
72 * }
73 * fprintf ("Recommended dB change for whole album: %+6.2f dB\n", GetAlbumGain() );
74 */
75
76 /*
77 * So here's the main source of potential code confusion:
78 *
79 * The filters applied to the incoming samples are IIR filters,
80 * meaning they rely on up to <filter order> number of previous samples
81 * AND up to <filter order> number of previous filtered samples.
82 *
83 * I set up the AnalyzeSamples routine to minimize memory usage and interface
84 * complexity. The speed isn't compromised too much (I don't think), but the
85 * internal complexity is higher than it should be for such a relatively
86 * simple routine.
87 *
88 * Optimization/clarity suggestions are welcome.
89 */
90
91 #ifdef HAVE_CONFIG_H
92 #include <config.h>
93 #endif
94
95 #include <stdio.h>
96 #include <stdlib.h>
97 #include <string.h>
98
99 #include "lame.h"
100 #include "machine.h"
101 #include "gain_analysis.h"
102
103 /* for each filter: */
104 /* [0] 48 kHz, [1] 44.1 kHz, [2] 32 kHz, [3] 24 kHz, [4] 22050 Hz, [5] 16 kHz, [6] 12 kHz, [7] is 11025 Hz, [8] 8 kHz */
105
106 #ifdef WIN32
107 #pragma warning ( disable : 4305 )
108 #endif
109
110
111 /*lint -save -e736 loss of precision */
112 static const Float_t ABYule[9][multiple_of(4, 2 * YULE_ORDER + 1)] = {
113 /* 20 18 16 14 12 10 8 6 4 2 0 19 17 15 13 11 9 7 5 3 1 */
114 { 0.00288463683916, 0.00012025322027, 0.00306428023191, 0.00594298065125, -0.02074045215285, 0.02161526843274, -0.01655260341619, -0.00009291677959, -0.00123395316851, -0.02160367184185, 0.03857599435200, 0.13919314567432, -0.86984376593551, 2.75465861874613, -5.87257861775999, 9.48293806319790,-12.28759895145294, 13.05504219327545,-11.34170355132042, 7.81501653005538, -3.84664617118067},
115 {-0.00187763777362, 0.00674613682247, -0.00240879051584, 0.01624864962975, -0.02596338512915, 0.02245293253339, -0.00834990904936, -0.00851165645469, -0.00848709379851, -0.02911007808948, 0.05418656406430, 0.13149317958808, -0.75104302451432, 2.19611684890774, -4.39470996079559, 6.85401540936998, -8.81498681370155, 9.47693607801280, -8.54751527471874, 6.36317777566148, -3.47845948550071},
116 {-0.00881362733839, 0.00651420667831, -0.01390589421898, 0.03174092540049, 0.00222312597743, 0.04781476674921, -0.05588393329856, 0.02163541888798, -0.06247880153653, -0.09331049056315, 0.15457299681924, 0.02347897407020, -0.05032077717131, 0.16378164858596, -0.45953458054983, 1.00595954808547, -1.67148153367602, 2.23697657451713, -2.64577170229825, 2.84868151156327, -2.37898834973084},
117 {-0.02950134983287, 0.00205861885564, -0.00000828086748, 0.06276101321749, -0.00584456039913, -0.02364141202522, -0.00915702933434, 0.03282930172664, -0.08587323730772, -0.22613988682123, 0.30296907319327, 0.00302439095741, 0.02005851806501, 0.04500235387352, -0.22138138954925, 0.39120800788284, -0.22638893773906, -0.16276719120440, -0.25656257754070, 1.07977492259970, -1.61273165137247},
118 {-0.01760176568150, -0.01635381384540, 0.00832043980773, 0.05724228140351, -0.00589500224440, -0.00469977914380, -0.07834489609479, 0.11921148675203, -0.11828570177555, -0.25572241425570, 0.33642304856132, 0.02977207319925, -0.04237348025746, 0.08333755284107, -0.04067510197014, -0.12453458140019, 0.47854794562326, -0.80774944671438, 0.12205022308084, 0.87350271418188, -1.49858979367799},
119 { 0.00541907748707, -0.03193428438915, -0.01863887810927, 0.10478503600251, 0.04097565135648, -0.12398163381748, 0.04078262797139, -0.01419140100551, -0.22784394429749, -0.14351757464547, 0.44915256608450, 0.03222754072173, 0.05784820375801, 0.06747620744683, 0.00613424350682, 0.22199650564824, -0.42029820170918, 0.00213767857124, -0.37256372942400, 0.29661783706366, -0.62820619233671},
120 {-0.00588215443421, -0.03788984554840, 0.08647503780351, 0.00647310677246, -0.27562961986224, 0.30931782841830, -0.18901604199609, 0.16744243493672, 0.16242137742230, -0.75464456939302, 0.56619470757641, 0.01807364323573, 0.01639907836189, -0.04784254229033, 0.06739368333110, -0.33032403314006, 0.45054734505008, 0.00819999645858, -0.26806001042947, 0.29156311971249, -1.04800335126349},
121 {-0.00749618797172, -0.03721611395801, 0.06920467763959, 0.01628462406333, -0.25344790059353, 0.15558449135573, 0.02377945217615, 0.17520704835522, -0.14289799034253, -0.53174909058578, 0.58100494960553, 0.01818801111503, 0.02442357316099, -0.02505961724053, -0.05246019024463, -0.23313271880868, 0.38952639978999, 0.14728154134330, -0.20256413484477, -0.31863563325245, -0.51035327095184},
122 {-0.02217936801134, 0.04788665548180, -0.04060034127000, -0.11202315195388, -0.02459864859345, 0.14590772289388, -0.10214864179676, 0.04267842219415, -0.00275953611929, -0.42163034350696, 0.53648789255105, 0.04704409688120, 0.05477720428674, -0.18823009262115, -0.17556493366449, 0.15113130533216, 0.26408300200955, -0.04678328784242, -0.03424681017675, -0.43193942311114, -0.25049871956020}
123 };
124
125 static const Float_t ABButter[9][multiple_of(4, 2 * BUTTER_ORDER + 1)] = {
126 /* 5 4 3 2 1 */
127 {0.98621192462708, 0.97261396931306, -1.97242384925416, -1.97223372919527, 0.98621192462708},
128 {0.98500175787242, 0.97022847566350, -1.97000351574484, -1.96977855582618, 0.98500175787242},
129 {0.97938932735214, 0.95920349965459, -1.95877865470428, -1.95835380975398, 0.97938932735214},
130 {0.97531843204928, 0.95124613669835, -1.95063686409857, -1.95002759149878, 0.97531843204928},
131 {0.97316523498161, 0.94705070426118, -1.94633046996323, -1.94561023566527, 0.97316523498161},
132 {0.96454515552826, 0.93034775234268, -1.92909031105652, -1.92783286977036, 0.96454515552826},
133 {0.96009142950541, 0.92177618768381, -1.92018285901082, -1.91858953033784, 0.96009142950541},
134 {0.95856916599601, 0.91885558323625, -1.91713833199203, -1.91542108074780, 0.95856916599601},
135 {0.94597685600279, 0.89487434461664, -1.89195371200558, -1.88903307939452, 0.94597685600279}
136 };
137
138 /*lint -restore */
139
140 #ifdef WIN32
141 #pragma warning ( default : 4305 )
142 #endif
143
144 /* When calling this procedure, make sure that ip[-order] and op[-order] point to real data! */
145
146 static void
filterYule(const Float_t * input,Float_t * output,size_t nSamples,const Float_t * const kernel)147 filterYule(const Float_t * input, Float_t * output, size_t nSamples, const Float_t * const kernel)
148 {
149 while (nSamples--) {
150 Float_t y0 = input[-10] * kernel[ 0];
151 Float_t y2 = input[ -9] * kernel[ 1];
152 Float_t y4 = input[ -8] * kernel[ 2];
153 Float_t y6 = input[ -7] * kernel[ 3];
154 Float_t s00 = y0 + y2 + y4 + y6;
155 Float_t y8 = input[ -6] * kernel[ 4];
156 Float_t yA = input[ -5] * kernel[ 5];
157 Float_t yC = input[ -4] * kernel[ 6];
158 Float_t yE = input[ -3] * kernel[ 7];
159 Float_t s01 = y8 + yA + yC + yE;
160 Float_t yG = input[ -2] * kernel[ 8] + input[ -1] * kernel[ 9];
161 Float_t yK = input[ 0] * kernel[10];
162
163 Float_t s1 = s00 + s01 + yG + yK;
164
165 Float_t x1 = output[-10] * kernel[11] + output[ -9] * kernel[12];
166 Float_t x5 = output[ -8] * kernel[13] + output[ -7] * kernel[14];
167 Float_t x9 = output[ -6] * kernel[15] + output[ -5] * kernel[16];
168 Float_t xD = output[ -4] * kernel[17] + output[ -3] * kernel[18];
169 Float_t xH = output[ -2] * kernel[19] + output[ -1] * kernel[20];
170
171 Float_t s2 = x1 + x5 + x9 + xD + xH;
172
173 output[0] = (Float_t)(s1 - s2);
174
175 ++output;
176 ++input;
177 }
178 }
179
180 static void
filterButter(const Float_t * input,Float_t * output,size_t nSamples,const Float_t * const kernel)181 filterButter(const Float_t * input, Float_t * output, size_t nSamples, const Float_t * const kernel)
182 {
183 while (nSamples--) {
184 Float_t s1 = input[-2] * kernel[0] + input[-1] * kernel[2] + input[ 0] * kernel[4];
185 Float_t s2 = output[-2] * kernel[1] + output[-1] * kernel[3];
186 output[0] = (Float_t)(s1 - s2);
187 ++output;
188 ++input;
189 }
190 }
191
192
193
194 static int ResetSampleFrequency(replaygain_t * rgData, long samplefreq);
195
196 /* returns a INIT_GAIN_ANALYSIS_OK if successful, INIT_GAIN_ANALYSIS_ERROR if not */
197
198 int
ResetSampleFrequency(replaygain_t * rgData,long samplefreq)199 ResetSampleFrequency(replaygain_t * rgData, long samplefreq)
200 {
201 /* zero out initial values, only first MAX_ORDER values */
202 memset(rgData->linprebuf, 0, MAX_ORDER * sizeof(*rgData->linprebuf));
203 memset(rgData->rinprebuf, 0, MAX_ORDER * sizeof(*rgData->rinprebuf));
204 memset(rgData->lstepbuf, 0, MAX_ORDER * sizeof(*rgData->lstepbuf));
205 memset(rgData->rstepbuf, 0, MAX_ORDER * sizeof(*rgData->rstepbuf));
206 memset(rgData->loutbuf, 0, MAX_ORDER * sizeof(*rgData->loutbuf));
207 memset(rgData->routbuf, 0, MAX_ORDER * sizeof(*rgData->routbuf));
208
209 switch ((int) (samplefreq)) {
210 case 48000:
211 rgData->freqindex = 0;
212 break;
213 case 44100:
214 rgData->freqindex = 1;
215 break;
216 case 32000:
217 rgData->freqindex = 2;
218 break;
219 case 24000:
220 rgData->freqindex = 3;
221 break;
222 case 22050:
223 rgData->freqindex = 4;
224 break;
225 case 16000:
226 rgData->freqindex = 5;
227 break;
228 case 12000:
229 rgData->freqindex = 6;
230 break;
231 case 11025:
232 rgData->freqindex = 7;
233 break;
234 case 8000:
235 rgData->freqindex = 8;
236 break;
237 default:
238 return INIT_GAIN_ANALYSIS_ERROR;
239 }
240
241 rgData->sampleWindow =
242 (samplefreq * RMS_WINDOW_TIME_NUMERATOR + RMS_WINDOW_TIME_DENOMINATOR -
243 1) / RMS_WINDOW_TIME_DENOMINATOR;
244
245 rgData->lsum = 0.;
246 rgData->rsum = 0.;
247 rgData->totsamp = 0;
248
249 memset(rgData->A, 0, sizeof(rgData->A));
250
251 return INIT_GAIN_ANALYSIS_OK;
252 }
253
254 int
InitGainAnalysis(replaygain_t * rgData,long samplefreq)255 InitGainAnalysis(replaygain_t * rgData, long samplefreq)
256 {
257 if (ResetSampleFrequency(rgData, samplefreq) != INIT_GAIN_ANALYSIS_OK) {
258 return INIT_GAIN_ANALYSIS_ERROR;
259 }
260
261 rgData->linpre = rgData->linprebuf + MAX_ORDER;
262 rgData->rinpre = rgData->rinprebuf + MAX_ORDER;
263 rgData->lstep = rgData->lstepbuf + MAX_ORDER;
264 rgData->rstep = rgData->rstepbuf + MAX_ORDER;
265 rgData->lout = rgData->loutbuf + MAX_ORDER;
266 rgData->rout = rgData->routbuf + MAX_ORDER;
267
268 memset(rgData->B, 0, sizeof(rgData->B));
269
270 return INIT_GAIN_ANALYSIS_OK;
271 }
272
273 /* returns GAIN_ANALYSIS_OK if successful, GAIN_ANALYSIS_ERROR if not */
274
275 int
AnalyzeSamples(replaygain_t * rgData,const Float_t * left_samples,const Float_t * right_samples,size_t num_samples,int num_channels)276 AnalyzeSamples(replaygain_t * rgData, const Float_t * left_samples, const Float_t * right_samples,
277 size_t num_samples, int num_channels)
278 {
279 const Float_t *curleft;
280 const Float_t *curright;
281 long batchsamples;
282 long cursamples;
283 long cursamplepos;
284 int i;
285 Float_t sum_l, sum_r;
286
287 if (num_samples == 0)
288 return GAIN_ANALYSIS_OK;
289
290 cursamplepos = 0;
291 batchsamples = (long) num_samples;
292
293 switch (num_channels) {
294 case 1:
295 right_samples = left_samples;
296 break;
297 case 2:
298 break;
299 default:
300 return GAIN_ANALYSIS_ERROR;
301 }
302
303 if (num_samples < MAX_ORDER) {
304 memcpy(rgData->linprebuf + MAX_ORDER, left_samples, num_samples * sizeof(Float_t));
305 memcpy(rgData->rinprebuf + MAX_ORDER, right_samples, num_samples * sizeof(Float_t));
306 }
307 else {
308 memcpy(rgData->linprebuf + MAX_ORDER, left_samples, MAX_ORDER * sizeof(Float_t));
309 memcpy(rgData->rinprebuf + MAX_ORDER, right_samples, MAX_ORDER * sizeof(Float_t));
310 }
311
312 while (batchsamples > 0) {
313 cursamples = batchsamples > rgData->sampleWindow - rgData->totsamp ?
314 rgData->sampleWindow - rgData->totsamp : batchsamples;
315 if (cursamplepos < MAX_ORDER) {
316 curleft = rgData->linpre + cursamplepos;
317 curright = rgData->rinpre + cursamplepos;
318 if (cursamples > MAX_ORDER - cursamplepos)
319 cursamples = MAX_ORDER - cursamplepos;
320 }
321 else {
322 curleft = left_samples + cursamplepos;
323 curright = right_samples + cursamplepos;
324 }
325
326 YULE_FILTER(curleft, rgData->lstep + rgData->totsamp, cursamples,
327 ABYule[rgData->freqindex]);
328 YULE_FILTER(curright, rgData->rstep + rgData->totsamp, cursamples,
329 ABYule[rgData->freqindex]);
330
331 BUTTER_FILTER(rgData->lstep + rgData->totsamp, rgData->lout + rgData->totsamp, cursamples,
332 ABButter[rgData->freqindex]);
333 BUTTER_FILTER(rgData->rstep + rgData->totsamp, rgData->rout + rgData->totsamp, cursamples,
334 ABButter[rgData->freqindex]);
335
336 curleft = rgData->lout + rgData->totsamp; /* Get the squared values */
337 curright = rgData->rout + rgData->totsamp;
338
339 sum_l = 0;
340 sum_r = 0;
341 i = cursamples & 0x03;
342 while (i--) {
343 Float_t const l = *curleft++;
344 Float_t const r = *curright++;
345 sum_l += l * l;
346 sum_r += r * r;
347 }
348 i = cursamples / 4;
349 while (i--) {
350 Float_t l0 = curleft[0] * curleft[0];
351 Float_t l1 = curleft[1] * curleft[1];
352 Float_t l2 = curleft[2] * curleft[2];
353 Float_t l3 = curleft[3] * curleft[3];
354 Float_t sl = l0 + l1 + l2 + l3;
355 Float_t r0 = curright[0] * curright[0];
356 Float_t r1 = curright[1] * curright[1];
357 Float_t r2 = curright[2] * curright[2];
358 Float_t r3 = curright[3] * curright[3];
359 Float_t sr = r0 + r1 + r2 + r3;
360 sum_l += sl;
361 curleft += 4;
362 sum_r += sr;
363 curright += 4;
364 }
365 rgData->lsum += sum_l;
366 rgData->rsum += sum_r;
367
368 batchsamples -= cursamples;
369 cursamplepos += cursamples;
370 rgData->totsamp += cursamples;
371 if (rgData->totsamp == rgData->sampleWindow) { /* Get the Root Mean Square (RMS) for this set of samples */
372 double const val =
373 STEPS_per_dB * 10. * log10((rgData->lsum + rgData->rsum) / rgData->totsamp * 0.5 +
374 1.e-37);
375 size_t ival = (val <= 0) ? 0 : (size_t) val;
376 if (ival >= sizeof(rgData->A) / sizeof(*(rgData->A)))
377 ival = sizeof(rgData->A) / sizeof(*(rgData->A)) - 1;
378 rgData->A[ival]++;
379 rgData->lsum = rgData->rsum = 0.;
380 memmove(rgData->loutbuf, rgData->loutbuf + rgData->totsamp,
381 MAX_ORDER * sizeof(Float_t));
382 memmove(rgData->routbuf, rgData->routbuf + rgData->totsamp,
383 MAX_ORDER * sizeof(Float_t));
384 memmove(rgData->lstepbuf, rgData->lstepbuf + rgData->totsamp,
385 MAX_ORDER * sizeof(Float_t));
386 memmove(rgData->rstepbuf, rgData->rstepbuf + rgData->totsamp,
387 MAX_ORDER * sizeof(Float_t));
388 rgData->totsamp = 0;
389 }
390 if (rgData->totsamp > rgData->sampleWindow) /* somehow I really screwed up: Error in programming! Contact author about totsamp > sampleWindow */
391 return GAIN_ANALYSIS_ERROR;
392 }
393 if (num_samples < MAX_ORDER) {
394 memmove(rgData->linprebuf, rgData->linprebuf + num_samples,
395 (MAX_ORDER - num_samples) * sizeof(Float_t));
396 memmove(rgData->rinprebuf, rgData->rinprebuf + num_samples,
397 (MAX_ORDER - num_samples) * sizeof(Float_t));
398 memcpy(rgData->linprebuf + MAX_ORDER - num_samples, left_samples,
399 num_samples * sizeof(Float_t));
400 memcpy(rgData->rinprebuf + MAX_ORDER - num_samples, right_samples,
401 num_samples * sizeof(Float_t));
402 }
403 else {
404 memcpy(rgData->linprebuf, left_samples + num_samples - MAX_ORDER,
405 MAX_ORDER * sizeof(Float_t));
406 memcpy(rgData->rinprebuf, right_samples + num_samples - MAX_ORDER,
407 MAX_ORDER * sizeof(Float_t));
408 }
409
410 return GAIN_ANALYSIS_OK;
411 }
412
413
414 static Float_t
analyzeResult(uint32_t const * Array,size_t len)415 analyzeResult(uint32_t const *Array, size_t len)
416 {
417 uint32_t elems;
418 uint32_t upper;
419 uint32_t sum;
420 size_t i;
421
422 elems = 0;
423 for (i = 0; i < len; i++)
424 elems += Array[i];
425 if (elems == 0)
426 return GAIN_NOT_ENOUGH_SAMPLES;
427
428 upper = (uint32_t) ceil(elems * (1. - RMS_PERCENTILE));
429 sum = 0;
430 for (i = len; i-- > 0;) {
431 sum += Array[i];
432 if (sum >= upper) {
433 break;
434 }
435 }
436
437 return (Float_t) ((Float_t) PINK_REF - (Float_t) i / (Float_t) STEPS_per_dB);
438 }
439
440
441 Float_t
GetTitleGain(replaygain_t * rgData)442 GetTitleGain(replaygain_t * rgData)
443 {
444 Float_t retval;
445 unsigned int i;
446
447 retval = analyzeResult(rgData->A, sizeof(rgData->A) / sizeof(*(rgData->A)));
448
449 for (i = 0; i < sizeof(rgData->A) / sizeof(*(rgData->A)); i++) {
450 rgData->B[i] += rgData->A[i];
451 rgData->A[i] = 0;
452 }
453
454 for (i = 0; i < MAX_ORDER; i++)
455 rgData->linprebuf[i] = rgData->lstepbuf[i]
456 = rgData->loutbuf[i]
457 = rgData->rinprebuf[i]
458 = rgData->rstepbuf[i]
459 = rgData->routbuf[i] = 0.f;
460
461 rgData->totsamp = 0;
462 rgData->lsum = rgData->rsum = 0.;
463 return retval;
464 }
465
466 #if 0
467 static Float_t GetAlbumGain(replaygain_t const* rgData);
468
469 Float_t
470 GetAlbumGain(replaygain_t const* rgData)
471 {
472 return analyzeResult(rgData->B, sizeof(rgData->B) / sizeof(*(rgData->B)));
473 }
474 #endif
475
476 /* end of gain_analysis.c */
477