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