configurable window size on volume normalisation, also set default of max_gain to...
[melted] / src / modules / core / filter_volume.c
1 /*
2 * filter_volume.c -- adjust audio volume
3 * Copyright (C) 2003-2004 Ushodaya Enterprises Limited
4 * Author: Dan Dennedy <dan@dennedy.org>
5 *
6 * This program is free software; you can redistribute it and/or modify
7 * it under the terms of the GNU General Public License as published by
8 * the Free Software Foundation; either version 2 of the License, or
9 * (at your option) any later version.
10 *
11 * This program 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
14 * GNU General Public License for more details.
15 *
16 * You should have received a copy of the GNU General Public License
17 * along with this program; if not, write to the Free Software Foundation,
18 * Inc., 59 Temple Place - Suite 330, Boston, MA 02111-1307, USA.
19 */
20
21 #include "filter_volume.h"
22
23 #include <framework/mlt_frame.h>
24
25 #include <stdio.h>
26 #include <stdlib.h>
27 #include <math.h>
28 #include <ctype.h>
29 #include <string.h>
30
31 #define MAX_CHANNELS 6
32 #define EPSILON 0.00001
33
34 /* The normalise functions come from the normalize utility:
35 Copyright (C) 1999--2002 Chris Vaill */
36
37 #define samp_width 16
38
39 #ifndef ROUND
40 # define ROUND(x) floor((x) + 0.5)
41 #endif
42
43 #define DBFSTOAMP(x) pow(10,(x)/20.0)
44
45 /** Return nonzero if the two strings are equal, ignoring case, up to
46 the first n characters.
47 */
48 int strncaseeq(const char *s1, const char *s2, size_t n)
49 {
50 for ( ; n > 0; n--)
51 {
52 if (tolower(*s1++) != tolower(*s2++))
53 return 0;
54 }
55 return 1;
56 }
57
58 /** Limiter function.
59
60 / tanh((x + lev) / (1-lev)) * (1-lev) - lev (for x < -lev)
61 |
62 x' = | x (for |x| <= lev)
63 |
64 \ tanh((x - lev) / (1-lev)) * (1-lev) + lev (for x > lev)
65
66 With limiter level = 0, this is equivalent to a tanh() function;
67 with limiter level = 1, this is equivalent to clipping.
68 */
69 static inline double limiter( double x, double lmtr_lvl )
70 {
71 double xp = x;
72
73 if (x < -lmtr_lvl)
74 xp = tanh((x + lmtr_lvl) / (1-lmtr_lvl)) * (1-lmtr_lvl) - lmtr_lvl;
75 else if (x > lmtr_lvl)
76 xp = tanh((x - lmtr_lvl) / (1-lmtr_lvl)) * (1-lmtr_lvl) + lmtr_lvl;
77
78 // if ( x != xp )
79 // fprintf( stderr, "filter_volume: sample %f limited %f\n", x, xp );
80
81 return xp;
82 }
83
84
85 /** Takes a full smoothing window, and returns the value of the center
86 element, smoothed.
87
88 Currently, just does a mean filter, but we could do a median or
89 gaussian filter here instead.
90 */
91 static inline double get_smoothed_data( double *buf, int count )
92 {
93 int i, j;
94 double smoothed = 0;
95
96 for ( i = 0, j = 0; i < count; i++ )
97 {
98 if ( buf[ i ] != -1.0 )
99 {
100 smoothed += buf[ i ];
101 j++;
102 }
103 }
104 smoothed /= j;
105 // fprintf( stderr, "smoothed over %d values, result %f\n", j, smoothed );
106
107 return smoothed;
108 }
109
110 /** Get the max power level (using RMS) and peak level of the audio segment.
111 */
112 double signal_max_power( int16_t *buffer, int channels, int samples, int16_t *peak )
113 {
114 // Determine numeric limits
115 int bytes_per_samp = (samp_width - 1) / 8 + 1;
116 int16_t max = (1 << (bytes_per_samp * 8 - 1)) - 1;
117 int16_t min = -max - 1;
118
119 double *sums = (double *) calloc( channels, sizeof(double) );
120 int c, i;
121 int16_t sample;
122 double pow, maxpow = 0;
123
124 /* initialize peaks to effectively -inf and +inf */
125 int16_t max_sample = min;
126 int16_t min_sample = max;
127
128 for ( i = 0; i < samples; i++ )
129 {
130 for ( c = 0; c < channels; c++ )
131 {
132 sample = *buffer++;
133 sums[ c ] += (double) sample * (double) sample;
134
135 /* track peak */
136 if ( sample > max_sample )
137 max_sample = sample;
138 else if ( sample < min_sample )
139 min_sample = sample;
140 }
141 }
142 for ( c = 0; c < channels; c++ )
143 {
144 pow = sums[ c ] / (double) samples;
145 if ( pow > maxpow )
146 maxpow = pow;
147 }
148
149 free( sums );
150
151 /* scale the pow value to be in the range 0.0 -- 1.0 */
152 maxpow /= ( (double) min * (double) min);
153
154 if ( -min_sample > max_sample )
155 *peak = min_sample / (double) min;
156 else
157 *peak = max_sample / (double) max;
158
159 return sqrt( maxpow );
160 }
161
162 /* ------ End normalize functions --------------------------------------- */
163
164 /** Get the audio.
165 */
166
167 static int filter_get_audio( mlt_frame frame, int16_t **buffer, mlt_audio_format *format, int *frequency, int *channels, int *samples )
168 {
169 // Get the properties of the a frame
170 mlt_properties properties = mlt_frame_properties( frame );
171 double gain = mlt_properties_get_double( properties, "gain" );
172 double max_gain = mlt_properties_get_double( properties, "volume.max_gain" );
173 double limiter_level = 0.5; /* -6 dBFS */
174 int normalise = mlt_properties_get_int( properties, "volume.normalise" );
175 double amplitude = mlt_properties_get_double( properties, "volume.amplitude" );
176 int i;
177 double sample;
178 int16_t peak;
179
180 if ( mlt_properties_get( properties, "volume.limiter" ) != NULL )
181 limiter_level = mlt_properties_get_double( properties, "volume.limiter" );
182
183 // Restore the original get_audio
184 frame->get_audio = mlt_properties_get_data( properties, "volume.get_audio", NULL );
185
186 // Get the producer's audio
187 mlt_frame_get_audio( frame, buffer, format, frequency, channels, samples );
188
189 // Determine numeric limits
190 int bytes_per_samp = (samp_width - 1) / 8 + 1;
191 int samplemax = (1 << (bytes_per_samp * 8 - 1)) - 1;
192 int samplemin = -samplemax - 1;
193
194 if ( normalise )
195 {
196 int window = mlt_properties_get_int( properties, "volume.window" );
197 double *smooth_buffer = mlt_properties_get_data( properties, "volume.smooth_buffer", NULL );
198 int *smooth_index = mlt_properties_get_data( properties, "volume.smooth_index", NULL );
199
200 if ( window > 0 && smooth_buffer != NULL )
201 {
202 // Compute the signal power and put into smoothing buffer
203 smooth_buffer[ *smooth_index ] = signal_max_power( *buffer, *channels, *samples, &peak );
204 // fprintf( stderr, "filter_volume: raw power %f ", smooth_buffer[ *smooth_index ] );
205 if ( smooth_buffer[ *smooth_index ] > EPSILON )
206 {
207 *smooth_index = ( *smooth_index + 1 ) % window;
208
209 // Smooth the data and compute the gain
210 //fprintf( stderr, "smoothed %f over %d frames\n", get_smoothed_data( smooth_buffer, window ), window );
211 gain *= amplitude / get_smoothed_data( smooth_buffer, window );
212 }
213 }
214 else
215 {
216 gain = amplitude / signal_max_power( *buffer, *channels, *samples, &peak );
217 }
218 }
219
220 // if ( gain > 1.0 && normalise )
221 // fprintf(stderr, "filter_volume: limiter level %f gain %f\n", limiter_level, gain );
222
223 if ( max_gain > 0 && gain > max_gain )
224 gain = max_gain;
225
226 // Apply the gain
227 for ( i = 0; i < ( *channels * *samples ); i++ )
228 {
229 sample = (*buffer)[i] * gain;
230 (*buffer)[i] = ROUND( sample );
231
232 if ( gain > 1.0 )
233 {
234 /* use limiter function instead of clipping */
235 if ( normalise )
236 (*buffer)[i] = ROUND( samplemax * limiter( sample / (double) samplemax, limiter_level ) );
237
238 /* perform clipping */
239 else if ( sample > samplemax )
240 (*buffer)[i] = samplemax;
241 else if ( sample < samplemin )
242 (*buffer)[i] = samplemin;
243 }
244 }
245
246 return 0;
247 }
248
249 /** Filter processing.
250 */
251
252 static mlt_frame filter_process( mlt_filter this, mlt_frame frame )
253 {
254 mlt_properties properties = mlt_frame_properties( frame );
255 mlt_properties filter_props = mlt_filter_properties( this );
256
257 // Propogate the gain property
258 if ( mlt_properties_get( properties, "gain" ) == NULL )
259 {
260 double gain = 1.0; // no adjustment
261
262 if ( mlt_properties_get( filter_props, "gain" ) != NULL )
263 {
264 char *p = mlt_properties_get( filter_props, "gain" );
265
266 if ( strncaseeq( p, "normalise", 9 ) )
267 mlt_properties_set( filter_props, "normalise", "" );
268 else
269 {
270 if ( strcmp( p, "" ) != 0 )
271 gain = fabs( strtod( p, &p) );
272
273 while ( isspace( *p ) )
274 p++;
275
276 /* check if "dB" is given after number */
277 if ( strncaseeq( p, "db", 2 ) )
278 gain = DBFSTOAMP( gain );
279 }
280 }
281 mlt_properties_set_double( properties, "gain", gain );
282 }
283
284 // Propogate the maximum gain property
285 if ( mlt_properties_get( filter_props, "max_gain" ) != NULL )
286 {
287 char *p = mlt_properties_get( filter_props, "max_gain" );
288 double gain = fabs( strtod( p, &p) ); // 0 = no max
289
290 while ( isspace( *p ) )
291 p++;
292
293 /* check if "dB" is given after number */
294 if ( strncaseeq( p, "db", 2 ) )
295 gain = DBFSTOAMP( gain );
296
297 mlt_properties_set_double( properties, "volume.max_gain", gain );
298 }
299
300 // Parse and propogate the limiter property
301 if ( mlt_properties_get( filter_props, "limiter" ) != NULL )
302 {
303 char *p = mlt_properties_get( filter_props, "limiter" );
304 double level = 0.5; /* -6dBFS */
305 if ( strcmp( p, "" ) != 0 )
306 level = strtod( p, &p);
307
308 while ( isspace( *p ) )
309 p++;
310
311 /* check if "dB" is given after number */
312 if ( strncaseeq( p, "db", 2 ) )
313 {
314 if ( level > 0 )
315 level = -level;
316 level = DBFSTOAMP( level );
317 }
318 else
319 {
320 if ( level < 0 )
321 level = -level;
322 }
323 mlt_properties_set_double( properties, "volume.limiter", level );
324 }
325
326 // Parse and propogate the normalise property
327 if ( mlt_properties_get( filter_props, "normalise" ) != NULL )
328 {
329 char *p = mlt_properties_get( filter_props, "normalise" );
330 double amplitude = 0.2511886431509580; /* -12dBFS */
331 if ( strcmp( p, "" ) != 0 )
332 amplitude = strtod( p, &p);
333
334 while ( isspace( *p ) )
335 p++;
336
337 /* check if "dB" is given after number */
338 if ( strncaseeq( p, "db", 2 ) )
339 {
340 if ( amplitude > 0 )
341 amplitude = -amplitude;
342 amplitude = DBFSTOAMP( amplitude );
343 }
344 else
345 {
346 if ( amplitude < 0 )
347 amplitude = -amplitude;
348 if ( amplitude > 1.0 )
349 amplitude = 1.0;
350 }
351 mlt_properties_set_int( properties, "volume.normalise", 1 );
352 mlt_properties_set_double( properties, "volume.amplitude", amplitude );
353 }
354
355 int window = mlt_properties_get_int( filter_props, "window" );
356 if ( mlt_properties_get( filter_props, "smooth_buffer" ) == NULL && window > 1 )
357 {
358 // Create a smoothing buffer for the calculated "max power" of frame of audio used in normalisation
359 double *smooth_buffer = (double*) calloc( window, sizeof( double ) );
360 int i;
361 for ( i = 0; i < window; i++ )
362 smooth_buffer[ i ] = -1.0;
363 mlt_properties_set_data( filter_props, "smooth_buffer", smooth_buffer, 0, free, NULL );
364 int *smooth_index = calloc( 1, sizeof( int ) );
365
366 mlt_properties_set_data( filter_props, "smooth_index", smooth_index, 0, free, NULL );
367 }
368
369 // Propogate the smoothing buffer properties
370 mlt_properties_set_int( properties, "volume.window", window );
371 mlt_properties_set_data( properties, "volume.smooth_buffer",
372 mlt_properties_get_data( filter_props, "smooth_buffer", NULL ), 0, NULL, NULL );
373 mlt_properties_set_data( properties, "volume.smooth_index",
374 mlt_properties_get_data( filter_props, "smooth_index", NULL ), 0, NULL, NULL );
375
376 // Backup the original get_audio (it's still needed)
377 mlt_properties_set_data( properties, "volume.get_audio", frame->get_audio, 0, NULL, NULL );
378
379 // Override the get_audio method
380 frame->get_audio = filter_get_audio;
381
382 return frame;
383 }
384
385 /** Constructor for the filter.
386 */
387
388 mlt_filter filter_volume_init( char *arg )
389 {
390 mlt_filter this = calloc( sizeof( struct mlt_filter_s ), 1 );
391 if ( this != NULL && mlt_filter_init( this, NULL ) == 0 )
392 {
393 mlt_properties properties = mlt_filter_properties( this );
394 this->process = filter_process;
395 if ( arg != NULL )
396 mlt_properties_set( properties, "gain", arg );
397
398 mlt_properties_set_int( properties, "window", 75 );
399 mlt_properties_set( properties, "max_gain", "20dB" );
400 }
401 return this;
402 }
403