1 /* GStreamer ReplayGain analysis
2 *
3 * Copyright (C) 2006 Rene Stadler <mail@renestadler.de>
4 * Copyright (C) 2001 David Robinson <David@Robinson.org>
5 * Glen Sawyer <glensawyer@hotmail.com>
6 *
7 * rganalysis.c: Analyze raw audio data in accordance with ReplayGain
8 *
9 * This library is free software; you can redistribute it and/or
10 * modify it under the terms of the GNU Lesser General Public License
11 * as published by the Free Software Foundation; either version 2.1 of
12 * the License, or (at your option) any later version.
13 *
14 * This library is distributed in the hope that it will be useful, but
15 * WITHOUT ANY WARRANTY; without even the implied warranty of
16 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
17 * Lesser General Public License for more details.
18 *
19 * You should have received a copy of the GNU Lesser General Public
20 * License along with this library; if not, write to the Free Software
21 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA
22 * 02110-1301 USA
23 */
24
25 /* Based on code with Copyright (C) 2001 David Robinson
26 * <David@Robinson.org> and Glen Sawyer <glensawyer@hotmail.com>,
27 * which is distributed under the LGPL as part of the vorbisgain
28 * program. The original code also mentions Frank Klemm
29 * (http://www.uni-jena.de/~pfk/mpp/) for having contributed lots of
30 * good code. Specifically, this is based on the file
31 * "gain_analysis.c" from vorbisgain version 0.34.
32 */
33
34 /* Room for future improvement: Mono data is currently in fact copied
35 * to two channels which get processed normally. This means that mono
36 * input data is processed twice.
37 */
38
39 /* Helpful information for understanding this code: The two IIR
40 * filters depend on previous input _and_ previous output samples (up
41 * to the filter's order number of samples). This explains the whole
42 * lot of memcpy'ing done in rg_analysis_analyze and why the context
43 * holds so many buffers.
44 */
45
46 #include <math.h>
47 #include <string.h>
48 #include <glib.h>
49
50 #include "rganalysis.h"
51
52 #define YULE_ORDER 10
53 #define BUTTER_ORDER 2
54 /* Percentile which is louder than the proposed level: */
55 #define RMS_PERCENTILE 95
56 /* Duration of RMS window in milliseconds: */
57 #define RMS_WINDOW_MSECS 50
58 /* Histogram array elements per dB: */
59 #define STEPS_PER_DB 100
60 /* Histogram upper bound in dB (normal max. values in the wild are
61 * assumed to be around 70, 80 dB): */
62 #define MAX_DB 120
63 /* Calibration value: */
64 #define PINK_REF 64.82 /* 298640883795 */
65
66 #define MAX_ORDER MAX (BUTTER_ORDER, YULE_ORDER)
67 #define MAX_SAMPLE_RATE 48000
68 /* The + 999 has the effect of ceil()ing: */
69 #define MAX_SAMPLE_WINDOW (guint) \
70 ((MAX_SAMPLE_RATE * RMS_WINDOW_MSECS + 999) / 1000)
71
72 /* Analysis result accumulator. */
73
74 struct _RgAnalysisAcc
75 {
76 guint32 histogram[STEPS_PER_DB * MAX_DB];
77 gdouble peak;
78 };
79
80 typedef struct _RgAnalysisAcc RgAnalysisAcc;
81
82 /* Analysis context. */
83
84 struct _RgAnalysisCtx
85 {
86 /* Filter buffers for left channel. */
87 gfloat inprebuf_l[MAX_ORDER * 2];
88 gfloat *inpre_l;
89 gfloat stepbuf_l[MAX_SAMPLE_WINDOW + MAX_ORDER];
90 gfloat *step_l;
91 gfloat outbuf_l[MAX_SAMPLE_WINDOW + MAX_ORDER];
92 gfloat *out_l;
93 /* Filter buffers for right channel. */
94 gfloat inprebuf_r[MAX_ORDER * 2];
95 gfloat *inpre_r;
96 gfloat stepbuf_r[MAX_SAMPLE_WINDOW + MAX_ORDER];
97 gfloat *step_r;
98 gfloat outbuf_r[MAX_SAMPLE_WINDOW + MAX_ORDER];
99 gfloat *out_r;
100
101 /* Number of samples to reach duration of the RMS window: */
102 guint window_n_samples;
103 /* Progress of the running window: */
104 guint window_n_samples_done;
105 gdouble window_square_sum;
106
107 gint sample_rate;
108 gint sample_rate_index;
109
110 RgAnalysisAcc track;
111 RgAnalysisAcc album;
112 void (*post_message) (gpointer analysis,
113 GstClockTime timestamp, GstClockTime duration, gdouble rglevel);
114 gpointer analysis;
115 /* The timestamp of the current incoming buffer. */
116 GstClockTime buffer_timestamp;
117 /* Number of samples processed in current buffer, during emit_signal,
118 this will always be on an RMS window boundary. */
119 guint buffer_n_samples_done;
120 };
121
122 /* Filter coefficients for the IIR filters that form the equal
123 * loudness filter. XFilter[ctx->sample_rate_index] gives the array
124 * of the X coefficients (A or B) for the configured sample rate. */
125
126 #ifdef _MSC_VER
127 /* Disable double-to-float warning: */
128 /* A better solution would be to append 'f' to each constant, but that
129 * makes the code ugly. */
130 #pragma warning ( disable : 4305 )
131 #endif
132
133 static const gfloat AYule[9][11] = {
134 {1., -3.84664617118067, 7.81501653005538, -11.34170355132042,
135 13.05504219327545, -12.28759895145294, 9.48293806319790,
136 -5.87257861775999, 2.75465861874613, -0.86984376593551,
137 0.13919314567432},
138 {1., -3.47845948550071, 6.36317777566148, -8.54751527471874, 9.47693607801280,
139 -8.81498681370155, 6.85401540936998, -4.39470996079559,
140 2.19611684890774, -0.75104302451432, 0.13149317958808},
141 {1., -2.37898834973084, 2.84868151156327, -2.64577170229825, 2.23697657451713,
142 -1.67148153367602, 1.00595954808547, -0.45953458054983,
143 0.16378164858596, -0.05032077717131, 0.02347897407020},
144 {1., -1.61273165137247, 1.07977492259970, -0.25656257754070,
145 -0.16276719120440, -0.22638893773906, 0.39120800788284,
146 -0.22138138954925, 0.04500235387352, 0.02005851806501,
147 0.00302439095741},
148 {1., -1.49858979367799, 0.87350271418188, 0.12205022308084, -0.80774944671438,
149 0.47854794562326, -0.12453458140019, -0.04067510197014,
150 0.08333755284107, -0.04237348025746, 0.02977207319925},
151 {1., -0.62820619233671, 0.29661783706366, -0.37256372942400, 0.00213767857124,
152 -0.42029820170918, 0.22199650564824, 0.00613424350682, 0.06747620744683,
153 0.05784820375801, 0.03222754072173},
154 {1., -1.04800335126349, 0.29156311971249, -0.26806001042947, 0.00819999645858,
155 0.45054734505008, -0.33032403314006, 0.06739368333110,
156 -0.04784254229033, 0.01639907836189, 0.01807364323573},
157 {1., -0.51035327095184, -0.31863563325245, -0.20256413484477,
158 0.14728154134330, 0.38952639978999, -0.23313271880868,
159 -0.05246019024463, -0.02505961724053, 0.02442357316099,
160 0.01818801111503},
161 {1., -0.25049871956020, -0.43193942311114, -0.03424681017675,
162 -0.04678328784242, 0.26408300200955, 0.15113130533216,
163 -0.17556493366449, -0.18823009262115, 0.05477720428674,
164 0.04704409688120}
165 };
166
167 static const gfloat BYule[9][11] = {
168 {0.03857599435200, -0.02160367184185, -0.00123395316851, -0.00009291677959,
169 -0.01655260341619, 0.02161526843274, -0.02074045215285,
170 0.00594298065125, 0.00306428023191, 0.00012025322027, 0.00288463683916},
171 {0.05418656406430, -0.02911007808948, -0.00848709379851, -0.00851165645469,
172 -0.00834990904936, 0.02245293253339, -0.02596338512915,
173 0.01624864962975, -0.00240879051584, 0.00674613682247,
174 -0.00187763777362},
175 {0.15457299681924, -0.09331049056315, -0.06247880153653, 0.02163541888798,
176 -0.05588393329856, 0.04781476674921, 0.00222312597743, 0.03174092540049,
177 -0.01390589421898, 0.00651420667831, -0.00881362733839},
178 {0.30296907319327, -0.22613988682123, -0.08587323730772, 0.03282930172664,
179 -0.00915702933434, -0.02364141202522, -0.00584456039913,
180 0.06276101321749, -0.00000828086748, 0.00205861885564,
181 -0.02950134983287},
182 {0.33642304856132, -0.25572241425570, -0.11828570177555, 0.11921148675203,
183 -0.07834489609479, -0.00469977914380, -0.00589500224440,
184 0.05724228140351, 0.00832043980773, -0.01635381384540,
185 -0.01760176568150},
186 {0.44915256608450, -0.14351757464547, -0.22784394429749, -0.01419140100551,
187 0.04078262797139, -0.12398163381748, 0.04097565135648, 0.10478503600251,
188 -0.01863887810927, -0.03193428438915, 0.00541907748707},
189 {0.56619470757641, -0.75464456939302, 0.16242137742230, 0.16744243493672,
190 -0.18901604199609, 0.30931782841830, -0.27562961986224,
191 0.00647310677246, 0.08647503780351, -0.03788984554840,
192 -0.00588215443421},
193 {0.58100494960553, -0.53174909058578, -0.14289799034253, 0.17520704835522,
194 0.02377945217615, 0.15558449135573, -0.25344790059353, 0.01628462406333,
195 0.06920467763959, -0.03721611395801, -0.00749618797172},
196 {0.53648789255105, -0.42163034350696, -0.00275953611929, 0.04267842219415,
197 -0.10214864179676, 0.14590772289388, -0.02459864859345,
198 -0.11202315195388, -0.04060034127000, 0.04788665548180,
199 -0.02217936801134}
200 };
201
202 static const gfloat AButter[9][3] = {
203 {1., -1.97223372919527, 0.97261396931306},
204 {1., -1.96977855582618, 0.97022847566350},
205 {1., -1.95835380975398, 0.95920349965459},
206 {1., -1.95002759149878, 0.95124613669835},
207 {1., -1.94561023566527, 0.94705070426118},
208 {1., -1.92783286977036, 0.93034775234268},
209 {1., -1.91858953033784, 0.92177618768381},
210 {1., -1.91542108074780, 0.91885558323625},
211 {1., -1.88903307939452, 0.89487434461664}
212 };
213
214 static const gfloat BButter[9][3] = {
215 {0.98621192462708, -1.97242384925416, 0.98621192462708},
216 {0.98500175787242, -1.97000351574484, 0.98500175787242},
217 {0.97938932735214, -1.95877865470428, 0.97938932735214},
218 {0.97531843204928, -1.95063686409857, 0.97531843204928},
219 {0.97316523498161, -1.94633046996323, 0.97316523498161},
220 {0.96454515552826, -1.92909031105652, 0.96454515552826},
221 {0.96009142950541, -1.92018285901082, 0.96009142950541},
222 {0.95856916599601, -1.91713833199203, 0.95856916599601},
223 {0.94597685600279, -1.89195371200558, 0.94597685600279}
224 };
225
226 #ifdef _MSC_VER
227 #pragma warning ( default : 4305 )
228 #endif
229
230 /* Filter functions. These access elements with negative indices of
231 * the input and output arrays (up to the filter's order). */
232
233 /* For much better performance, the function below has been
234 * implemented by unrolling the inner loop for our two use cases. */
235
236 /*
237 * static inline void
238 * apply_filter (const gfloat * input, gfloat * output, guint n_samples,
239 * const gfloat * a, const gfloat * b, guint order)
240 * {
241 * gfloat y;
242 * gint i, k;
243 *
244 * for (i = 0; i < n_samples; i++) {
245 * y = input[i] * b[0];
246 * for (k = 1; k <= order; k++)
247 * y += input[i - k] * b[k] - output[i - k] * a[k];
248 * output[i] = y;
249 * }
250 * }
251 */
252
253 static inline void
yule_filter(const gfloat * input,gfloat * output,const gfloat * a,const gfloat * b)254 yule_filter (const gfloat * input, gfloat * output,
255 const gfloat * a, const gfloat * b)
256 {
257 /* 1e-10 is added below to avoid running into denormals when operating on
258 * near silence. */
259
260 output[0] = 1e-10 + input[0] * b[0]
261 + input[-1] * b[1] - output[-1] * a[1]
262 + input[-2] * b[2] - output[-2] * a[2]
263 + input[-3] * b[3] - output[-3] * a[3]
264 + input[-4] * b[4] - output[-4] * a[4]
265 + input[-5] * b[5] - output[-5] * a[5]
266 + input[-6] * b[6] - output[-6] * a[6]
267 + input[-7] * b[7] - output[-7] * a[7]
268 + input[-8] * b[8] - output[-8] * a[8]
269 + input[-9] * b[9] - output[-9] * a[9]
270 + input[-10] * b[10] - output[-10] * a[10];
271 }
272
273 static inline void
butter_filter(const gfloat * input,gfloat * output,const gfloat * a,const gfloat * b)274 butter_filter (const gfloat * input, gfloat * output,
275 const gfloat * a, const gfloat * b)
276 {
277 output[0] = input[0] * b[0]
278 + input[-1] * b[1] - output[-1] * a[1]
279 + input[-2] * b[2] - output[-2] * a[2];
280 }
281
282 /* Because butter_filter and yule_filter are inlined, this function is
283 * a bit blown-up (code-size wise), but not inlining gives a ca. 40%
284 * performance penalty. */
285
286 static inline void
apply_filters(const RgAnalysisCtx * ctx,const gfloat * input_l,const gfloat * input_r,guint n_samples)287 apply_filters (const RgAnalysisCtx * ctx, const gfloat * input_l,
288 const gfloat * input_r, guint n_samples)
289 {
290 const gfloat *ayule = AYule[ctx->sample_rate_index];
291 const gfloat *byule = BYule[ctx->sample_rate_index];
292 const gfloat *abutter = AButter[ctx->sample_rate_index];
293 const gfloat *bbutter = BButter[ctx->sample_rate_index];
294 gint pos = ctx->window_n_samples_done;
295 gint i;
296
297 for (i = 0; i < n_samples; i++, pos++) {
298 yule_filter (input_l + i, ctx->step_l + pos, ayule, byule);
299 butter_filter (ctx->step_l + pos, ctx->out_l + pos, abutter, bbutter);
300
301 yule_filter (input_r + i, ctx->step_r + pos, ayule, byule);
302 butter_filter (ctx->step_r + pos, ctx->out_r + pos, abutter, bbutter);
303 }
304 }
305
306 /* Clear filter buffer state and current RMS window. */
307
308 static void
reset_filters(RgAnalysisCtx * ctx)309 reset_filters (RgAnalysisCtx * ctx)
310 {
311 gint i;
312
313 for (i = 0; i < MAX_ORDER; i++) {
314
315 ctx->inprebuf_l[i] = 0.;
316 ctx->stepbuf_l[i] = 0.;
317 ctx->outbuf_l[i] = 0.;
318
319 ctx->inprebuf_r[i] = 0.;
320 ctx->stepbuf_r[i] = 0.;
321 ctx->outbuf_r[i] = 0.;
322 }
323
324 ctx->window_square_sum = 0.;
325 ctx->window_n_samples_done = 0;
326 }
327
328 /* Accumulator functions. */
329
330 /* Add two accumulators in-place. The sum is defined as the result of
331 * the vector sum of the histogram array and the maximum value of the
332 * peak field. Thus "adding" the accumulators for all tracks yields
333 * the correct result for obtaining the album gain and peak. */
334
335 static void
accumulator_add(RgAnalysisAcc * acc,const RgAnalysisAcc * acc_other)336 accumulator_add (RgAnalysisAcc * acc, const RgAnalysisAcc * acc_other)
337 {
338 gint i;
339
340 for (i = 0; i < G_N_ELEMENTS (acc->histogram); i++)
341 acc->histogram[i] += acc_other->histogram[i];
342
343 acc->peak = MAX (acc->peak, acc_other->peak);
344 }
345
346 /* Reset an accumulator to zero. */
347
348 static void
accumulator_clear(RgAnalysisAcc * acc)349 accumulator_clear (RgAnalysisAcc * acc)
350 {
351 memset (acc->histogram, 0, sizeof (acc->histogram));
352 acc->peak = 0.;
353 }
354
355 /* Obtain final analysis result from an accumulator. Returns TRUE on
356 * success, FALSE on error (if accumulator is still zero). */
357
358 static gboolean
accumulator_result(const RgAnalysisAcc * acc,gdouble * result_gain,gdouble * result_peak)359 accumulator_result (const RgAnalysisAcc * acc, gdouble * result_gain,
360 gdouble * result_peak)
361 {
362 guint32 sum = 0;
363 guint32 upper;
364 guint i;
365
366 for (i = 0; i < G_N_ELEMENTS (acc->histogram); i++)
367 sum += acc->histogram[i];
368
369 if (sum == 0)
370 /* All entries are 0: We got less than 50ms of data. */
371 return FALSE;
372
373 upper = (guint32) ceil (sum * (1. - (gdouble) (RMS_PERCENTILE / 100.)));
374
375 for (i = G_N_ELEMENTS (acc->histogram); i--;) {
376 if (upper <= acc->histogram[i])
377 break;
378 upper -= acc->histogram[i];
379 }
380
381 if (result_peak != NULL)
382 *result_peak = acc->peak;
383 if (result_gain != NULL)
384 *result_gain = PINK_REF - (gdouble) i / STEPS_PER_DB;
385
386 return TRUE;
387 }
388
389 /* Functions that operate on contexts, for external usage. */
390
391 /* Create a new context. Before it can be used, a sample rate must be
392 * configured using rg_analysis_set_sample_rate. */
393
394 RgAnalysisCtx *
rg_analysis_new(void)395 rg_analysis_new (void)
396 {
397 RgAnalysisCtx *ctx;
398
399 ctx = g_new (RgAnalysisCtx, 1);
400
401 ctx->inpre_l = ctx->inprebuf_l + MAX_ORDER;
402 ctx->step_l = ctx->stepbuf_l + MAX_ORDER;
403 ctx->out_l = ctx->outbuf_l + MAX_ORDER;
404
405 ctx->inpre_r = ctx->inprebuf_r + MAX_ORDER;
406 ctx->step_r = ctx->stepbuf_r + MAX_ORDER;
407 ctx->out_r = ctx->outbuf_r + MAX_ORDER;
408
409 ctx->sample_rate = 0;
410
411 accumulator_clear (&ctx->track);
412 accumulator_clear (&ctx->album);
413
414 return ctx;
415 }
416
417 static void
reset_silence_detection(RgAnalysisCtx * ctx)418 reset_silence_detection (RgAnalysisCtx * ctx)
419 {
420 ctx->buffer_timestamp = GST_CLOCK_TIME_NONE;
421 ctx->buffer_n_samples_done = 0;
422 }
423
424 /* Adapt to given sample rate. Does nothing if already the current
425 * rate (returns TRUE then). Returns FALSE only if given sample rate
426 * is not supported. If the configured rate changes, the last
427 * unprocessed incomplete 50ms chunk of data is dropped because the
428 * filters are reset. */
429
430 gboolean
rg_analysis_set_sample_rate(RgAnalysisCtx * ctx,gint sample_rate)431 rg_analysis_set_sample_rate (RgAnalysisCtx * ctx, gint sample_rate)
432 {
433 g_return_val_if_fail (ctx != NULL, FALSE);
434
435 if (ctx->sample_rate == sample_rate)
436 return TRUE;
437
438 switch (sample_rate) {
439 case 48000:
440 ctx->sample_rate_index = 0;
441 break;
442 case 44100:
443 ctx->sample_rate_index = 1;
444 break;
445 case 32000:
446 ctx->sample_rate_index = 2;
447 break;
448 case 24000:
449 ctx->sample_rate_index = 3;
450 break;
451 case 22050:
452 ctx->sample_rate_index = 4;
453 break;
454 case 16000:
455 ctx->sample_rate_index = 5;
456 break;
457 case 12000:
458 ctx->sample_rate_index = 6;
459 break;
460 case 11025:
461 ctx->sample_rate_index = 7;
462 break;
463 case 8000:
464 ctx->sample_rate_index = 8;
465 break;
466 default:
467 return FALSE;
468 }
469
470 ctx->sample_rate = sample_rate;
471 /* The + 999 has the effect of ceil()ing: */
472 ctx->window_n_samples = (guint) ((sample_rate * RMS_WINDOW_MSECS + 999)
473 / 1000);
474
475 reset_filters (ctx);
476 reset_silence_detection (ctx);
477
478 return TRUE;
479 }
480
481 void
rg_analysis_init_silence_detection(RgAnalysisCtx * ctx,void (* post_message)(gpointer analysis,GstClockTime timestamp,GstClockTime duration,gdouble rglevel),gpointer analysis)482 rg_analysis_init_silence_detection (RgAnalysisCtx * ctx,
483 void (*post_message) (gpointer analysis, GstClockTime timestamp,
484 GstClockTime duration, gdouble rglevel), gpointer analysis)
485 {
486 ctx->post_message = post_message;
487 ctx->analysis = analysis;
488 reset_silence_detection (ctx);
489 }
490
491 void
rg_analysis_start_buffer(RgAnalysisCtx * ctx,GstClockTime buffer_timestamp)492 rg_analysis_start_buffer (RgAnalysisCtx * ctx, GstClockTime buffer_timestamp)
493 {
494 ctx->buffer_timestamp = buffer_timestamp;
495 ctx->buffer_n_samples_done = 0;
496 }
497
498 void
rg_analysis_destroy(RgAnalysisCtx * ctx)499 rg_analysis_destroy (RgAnalysisCtx * ctx)
500 {
501 g_free (ctx);
502 }
503
504 /* Entry points for analyzing sample data in common raw data formats.
505 * The stereo format functions expect interleaved frames. It is
506 * possible to pass data in different formats for the same context,
507 * there are no restrictions. All functions have the same signature;
508 * the depth argument for the float functions is not variable and must
509 * be given the value 32. */
510
511 void
rg_analysis_analyze_mono_float(RgAnalysisCtx * ctx,gconstpointer data,gsize size,guint depth)512 rg_analysis_analyze_mono_float (RgAnalysisCtx * ctx, gconstpointer data,
513 gsize size, guint depth)
514 {
515 gfloat conv_samples[512];
516 const gfloat *samples = (gfloat *) data;
517 guint n_samples = size / sizeof (gfloat);
518 gint i;
519
520 g_return_if_fail (depth == 32);
521 g_return_if_fail (size % sizeof (gfloat) == 0);
522
523 while (n_samples) {
524 gint n = MIN (n_samples, G_N_ELEMENTS (conv_samples));
525
526 n_samples -= n;
527 memcpy (conv_samples, samples, n * sizeof (gfloat));
528 for (i = 0; i < n; i++) {
529 ctx->track.peak = MAX (ctx->track.peak, fabs (conv_samples[i]));
530 conv_samples[i] *= 32768.;
531 }
532 samples += n;
533 rg_analysis_analyze (ctx, conv_samples, NULL, n);
534 }
535 }
536
537 void
rg_analysis_analyze_stereo_float(RgAnalysisCtx * ctx,gconstpointer data,gsize size,guint depth)538 rg_analysis_analyze_stereo_float (RgAnalysisCtx * ctx, gconstpointer data,
539 gsize size, guint depth)
540 {
541 gfloat conv_samples_l[256];
542 gfloat conv_samples_r[256];
543 const gfloat *samples = (gfloat *) data;
544 guint n_frames = size / (sizeof (gfloat) * 2);
545 gint i;
546
547 g_return_if_fail (depth == 32);
548 g_return_if_fail (size % (sizeof (gfloat) * 2) == 0);
549
550 while (n_frames) {
551 gint n = MIN (n_frames, G_N_ELEMENTS (conv_samples_l));
552
553 n_frames -= n;
554 for (i = 0; i < n; i++) {
555 gfloat old_sample;
556
557 old_sample = samples[2 * i];
558 ctx->track.peak = MAX (ctx->track.peak, fabs (old_sample));
559 conv_samples_l[i] = old_sample * 32768.;
560
561 old_sample = samples[2 * i + 1];
562 ctx->track.peak = MAX (ctx->track.peak, fabs (old_sample));
563 conv_samples_r[i] = old_sample * 32768.;
564 }
565 samples += 2 * n;
566 rg_analysis_analyze (ctx, conv_samples_l, conv_samples_r, n);
567 }
568 }
569
570 void
rg_analysis_analyze_mono_int16(RgAnalysisCtx * ctx,gconstpointer data,gsize size,guint depth)571 rg_analysis_analyze_mono_int16 (RgAnalysisCtx * ctx, gconstpointer data,
572 gsize size, guint depth)
573 {
574 gfloat conv_samples[512];
575 gint32 peak_sample = 0;
576 const gint16 *samples = (gint16 *) data;
577 guint n_samples = size / sizeof (gint16);
578 gint shift = 1 << (sizeof (gint16) * 8 - depth);
579 gint i;
580
581 g_return_if_fail (depth <= (sizeof (gint16) * 8));
582 g_return_if_fail (size % sizeof (gint16) == 0);
583
584 while (n_samples) {
585 gint n = MIN (n_samples, G_N_ELEMENTS (conv_samples));
586
587 n_samples -= n;
588 for (i = 0; i < n; i++) {
589 gint16 old_sample = samples[i] * shift;
590
591 peak_sample = MAX (peak_sample, ABS ((gint32) old_sample));
592 conv_samples[i] = (gfloat) old_sample;
593 }
594 samples += n;
595 rg_analysis_analyze (ctx, conv_samples, NULL, n);
596 }
597 ctx->track.peak = MAX (ctx->track.peak,
598 (gdouble) peak_sample / ((gdouble) (1u << 15)));
599 }
600
601 void
rg_analysis_analyze_stereo_int16(RgAnalysisCtx * ctx,gconstpointer data,gsize size,guint depth)602 rg_analysis_analyze_stereo_int16 (RgAnalysisCtx * ctx, gconstpointer data,
603 gsize size, guint depth)
604 {
605 gfloat conv_samples_l[256];
606 gfloat conv_samples_r[256];
607 gint32 peak_sample = 0;
608 const gint16 *samples = (gint16 *) data;
609 guint n_frames = size / (sizeof (gint16) * 2);
610 gint shift = 1 << (sizeof (gint16) * 8 - depth);
611 gint i;
612
613 g_return_if_fail (depth <= (sizeof (gint16) * 8));
614 g_return_if_fail (size % (sizeof (gint16) * 2) == 0);
615
616 while (n_frames) {
617 gint n = MIN (n_frames, G_N_ELEMENTS (conv_samples_l));
618
619 n_frames -= n;
620 for (i = 0; i < n; i++) {
621 gint16 old_sample;
622
623 old_sample = samples[2 * i] * shift;
624 peak_sample = MAX (peak_sample, ABS ((gint32) old_sample));
625 conv_samples_l[i] = (gfloat) old_sample;
626
627 old_sample = samples[2 * i + 1] * shift;
628 peak_sample = MAX (peak_sample, ABS ((gint32) old_sample));
629 conv_samples_r[i] = (gfloat) old_sample;
630 }
631 samples += 2 * n;
632 rg_analysis_analyze (ctx, conv_samples_l, conv_samples_r, n);
633 }
634 ctx->track.peak = MAX (ctx->track.peak,
635 (gdouble) peak_sample / ((gdouble) (1u << 15)));
636 }
637
638 /* Analyze the given chunk of samples. The sample data is given in
639 * floating point format but should be scaled such that the values
640 * +/-32768.0 correspond to the -0dBFS reference amplitude.
641 *
642 * samples_l: Buffer with sample data for the left channel or of the
643 * mono channel.
644 *
645 * samples_r: Buffer with sample data for the right channel or NULL
646 * for mono.
647 *
648 * n_samples: Number of samples passed in each buffer.
649 */
650
651 void
rg_analysis_analyze(RgAnalysisCtx * ctx,const gfloat * samples_l,const gfloat * samples_r,guint n_samples)652 rg_analysis_analyze (RgAnalysisCtx * ctx, const gfloat * samples_l,
653 const gfloat * samples_r, guint n_samples)
654 {
655 const gfloat *input_l, *input_r;
656 guint n_samples_done;
657 gint i;
658
659 g_return_if_fail (ctx != NULL);
660 g_return_if_fail (samples_l != NULL);
661 g_return_if_fail (ctx->sample_rate != 0);
662
663 if (n_samples == 0)
664 return;
665
666 if (samples_r == NULL)
667 /* Mono. */
668 samples_r = samples_l;
669
670 memcpy (ctx->inpre_l, samples_l,
671 MIN (n_samples, MAX_ORDER) * sizeof (gfloat));
672 memcpy (ctx->inpre_r, samples_r,
673 MIN (n_samples, MAX_ORDER) * sizeof (gfloat));
674
675 n_samples_done = 0;
676 while (n_samples_done < n_samples) {
677 /* Limit number of samples to be processed in this iteration to
678 * the number needed to complete the next window: */
679 guint n_samples_current = MIN (n_samples - n_samples_done,
680 ctx->window_n_samples - ctx->window_n_samples_done);
681
682 if (n_samples_done < MAX_ORDER) {
683 input_l = ctx->inpre_l + n_samples_done;
684 input_r = ctx->inpre_r + n_samples_done;
685 n_samples_current = MIN (n_samples_current, MAX_ORDER - n_samples_done);
686 } else {
687 input_l = samples_l + n_samples_done;
688 input_r = samples_r + n_samples_done;
689 }
690
691 apply_filters (ctx, input_l, input_r, n_samples_current);
692
693 /* Update the square sum. */
694 for (i = 0; i < n_samples_current; i++)
695 ctx->window_square_sum += ctx->out_l[ctx->window_n_samples_done + i]
696 * ctx->out_l[ctx->window_n_samples_done + i]
697 + ctx->out_r[ctx->window_n_samples_done + i]
698 * ctx->out_r[ctx->window_n_samples_done + i];
699
700 ctx->window_n_samples_done += n_samples_current;
701 ctx->buffer_n_samples_done += n_samples_current;
702
703 g_return_if_fail (ctx->window_n_samples_done <= ctx->window_n_samples);
704
705 if (ctx->window_n_samples_done == ctx->window_n_samples) {
706 /* Get the Root Mean Square (RMS) for this set of samples. */
707 gdouble val = STEPS_PER_DB * 10. * log10 (ctx->window_square_sum /
708 ctx->window_n_samples * 0.5 + 1.e-37);
709 gint ival = CLAMP ((gint) val, 0,
710 (gint) G_N_ELEMENTS (ctx->track.histogram) - 1);
711 /* Compute the per-window gain */
712 const gdouble gain = PINK_REF - (gdouble) ival / STEPS_PER_DB;
713 const GstClockTime timestamp = ctx->buffer_timestamp
714 + gst_util_uint64_scale_int_ceil (GST_SECOND,
715 ctx->buffer_n_samples_done,
716 ctx->sample_rate)
717 - RMS_WINDOW_MSECS * GST_MSECOND;
718
719 ctx->post_message (ctx->analysis, timestamp,
720 RMS_WINDOW_MSECS * GST_MSECOND, -gain);
721
722
723 ctx->track.histogram[ival]++;
724 ctx->window_square_sum = 0.;
725 ctx->window_n_samples_done = 0;
726
727 /* No need for memmove here, the areas never overlap: Even for
728 * the smallest sample rate, the number of samples needed for
729 * the window is greater than MAX_ORDER. */
730
731 memcpy (ctx->stepbuf_l, ctx->stepbuf_l + ctx->window_n_samples,
732 MAX_ORDER * sizeof (gfloat));
733 memcpy (ctx->outbuf_l, ctx->outbuf_l + ctx->window_n_samples,
734 MAX_ORDER * sizeof (gfloat));
735
736 memcpy (ctx->stepbuf_r, ctx->stepbuf_r + ctx->window_n_samples,
737 MAX_ORDER * sizeof (gfloat));
738 memcpy (ctx->outbuf_r, ctx->outbuf_r + ctx->window_n_samples,
739 MAX_ORDER * sizeof (gfloat));
740 }
741
742 n_samples_done += n_samples_current;
743 }
744
745 if (n_samples >= MAX_ORDER) {
746
747 memcpy (ctx->inprebuf_l, samples_l + n_samples - MAX_ORDER,
748 MAX_ORDER * sizeof (gfloat));
749
750 memcpy (ctx->inprebuf_r, samples_r + n_samples - MAX_ORDER,
751 MAX_ORDER * sizeof (gfloat));
752
753 } else {
754
755 memmove (ctx->inprebuf_l, ctx->inprebuf_l + n_samples,
756 (MAX_ORDER - n_samples) * sizeof (gfloat));
757 memcpy (ctx->inprebuf_l + MAX_ORDER - n_samples, samples_l,
758 n_samples * sizeof (gfloat));
759
760 memmove (ctx->inprebuf_r, ctx->inprebuf_r + n_samples,
761 (MAX_ORDER - n_samples) * sizeof (gfloat));
762 memcpy (ctx->inprebuf_r + MAX_ORDER - n_samples, samples_r,
763 n_samples * sizeof (gfloat));
764
765 }
766 }
767
768 /* Obtain track gain and peak. Returns TRUE on success. Can fail if
769 * not enough samples have been processed. Updates album accumulator.
770 * Resets track accumulator. */
771
772 gboolean
rg_analysis_track_result(RgAnalysisCtx * ctx,gdouble * gain,gdouble * peak)773 rg_analysis_track_result (RgAnalysisCtx * ctx, gdouble * gain, gdouble * peak)
774 {
775 gboolean result;
776
777 g_return_val_if_fail (ctx != NULL, FALSE);
778
779 accumulator_add (&ctx->album, &ctx->track);
780 result = accumulator_result (&ctx->track, gain, peak);
781 accumulator_clear (&ctx->track);
782
783 reset_filters (ctx);
784 reset_silence_detection (ctx);
785
786 return result;
787 }
788
789 /* Obtain album gain and peak. Returns TRUE on success. Can fail if
790 * not enough samples have been processed. Resets album
791 * accumulator. */
792
793 gboolean
rg_analysis_album_result(RgAnalysisCtx * ctx,gdouble * gain,gdouble * peak)794 rg_analysis_album_result (RgAnalysisCtx * ctx, gdouble * gain, gdouble * peak)
795 {
796 gboolean result;
797
798 g_return_val_if_fail (ctx != NULL, FALSE);
799
800 result = accumulator_result (&ctx->album, gain, peak);
801 accumulator_clear (&ctx->album);
802
803 return result;
804 }
805
806 void
rg_analysis_reset_album(RgAnalysisCtx * ctx)807 rg_analysis_reset_album (RgAnalysisCtx * ctx)
808 {
809 accumulator_clear (&ctx->album);
810 }
811
812 /* Reset internal buffers as well as track and album accumulators.
813 * Configured sample rate is kept intact. */
814
815 void
rg_analysis_reset(RgAnalysisCtx * ctx)816 rg_analysis_reset (RgAnalysisCtx * ctx)
817 {
818 g_return_if_fail (ctx != NULL);
819
820 reset_filters (ctx);
821 accumulator_clear (&ctx->track);
822 accumulator_clear (&ctx->album);
823 reset_silence_detection (ctx);
824 }
825