avfilter/af_afir: add irnorm and irlink options

Deprecate gtype option.

Paul B Mahol committed 2 years ago

Commit 5452cbd1 parent 07c303b4 files · +91 −119Browse files

Unified · Split

4 files

doc/filters.texi +14 −23
libavfilter/af_afir.c +47 −10
libavfilter/af_afir.h +3 −0
libavfilter/afir_template.c +27 −86

@@ -1794,33 +1794,24 @@ Set wet gain. This sets final output gain. · 1793 unmodified lines
1794	1794	Set Impulse Response filter length. Default is 1, which means whole IR is processed.
1795	1795
1796	1796	@item gtype
1797		−Enable applying gain measured from power of IR.
	1797	+This option is deprecated, and does nothing.
1798	1798
1799		−Set which approach to use for auto gain measurement.
	1799	+@item irnorm
	1800	+Set norm to be applied to IR coefficients before filtering.
	1801	+Allowed range is from @var{-1} to @var{2}.
	1802	+IR coefficients are normalized with calculated vector norm set by this option.
	1803	+For negative values, no norm is calculated, and IR coefficients are not modified at all.
	1804	+Default is @var{1}.
1800	1805
1801		−@table @option
1802		−@item none
1803		−Do not apply any gain.
1804		−
1805		−@item peak
1806		−select peak gain, very conservative approach. This is default value.
1807		−
1808		−@item dc
1809		−select DC gain, limited application.
1810		−
1811		−@item gn
1812		−select gain to noise approach, this is most popular one.
1813		−
1814		−@item ac
1815		−select AC gain.
1816		−
1817		−@item rms
1818		−select RMS gain.
1819		−@end table
	1806	+@item irlink
	1807	+For multichannel IR if this option is set to @var{true}. All IR channels will be
	1808	+normalized with maximal measured gain of all IR channels coefficients as set by @code{irnorm} option.
	1809	+When disabled, all IR coefficients in each IR channel will be normalized independently.
	1810	+Default is @var{true}.
1820	1811
1821	1812	@item irgain
1822	1813	Set gain to be applied to IR coefficients before filtering.
1823		−Allowed range is 0 to 1. This gain is applied after any gain applied with @var{gtype} option.
	1814	+Allowed range is 0 to 1. This gain is applied after any gain applied with @var{irnorm} option.
1824	1815
1825	1816	@item irfmt
1826	1817	Set format of IR stream. Can be @code{mono} or @code{input}.
@@ -1899,7 +1890,7 @@ ffmpeg -i input.wav -i middle_tunnel_1way_mono.wav -lavfi afir output.wav · 72 unmodified lines
1899	1890	Apply true stereo processing given input stereo stream, and two stereo impulse responses for left and right channel,
1900	1891	the impulse response files are files with names l_ir.wav and r_ir.wav:
1901	1892	@example
1902		−"pan=4C\|c0=FL\|c1=FL\|c2=FR\|c3=FR[a];amovie=l_ir.wav[LIR];amovie=r_ir.wav[RIR];[LIR][RIR]amerge[ir];[a][ir]afir=irfmt=input:gtype=gn:irgain=-5dB,pan=stereo\|FL<c0+c2\|FR<c1+c3"
	1893	+"pan=4C\|c0=FL\|c1=FL\|c2=FR\|c3=FR[a];amovie=l_ir.wav[LIR];amovie=r_ir.wav[RIR];[LIR][RIR]amerge[ir];[a][ir]afir=irfmt=input:irgain=-5dB,pan=stereo\|FL<c0+c2\|FR<c1+c3"
1903	1894	@end example
1904	1895	@end itemize
1905	1896

libavfilter/af_afir.c+47−10

@@ -393,6 +393,22 @@ skip: · 392 unmodified lines
393	393
394	394	switch (s->format) {
395	395	case AV_SAMPLE_FMT_FLTP:
	396	+ for (int ch = 0; ch < s->nb_channels; ch++) {
	397	+ const float tsrc = (const float )s->ir[selir]->extended_data[!s->one2many * ch];
	398	+
	399	+ s->ch_gain[ch] = ir_gain_float(ctx, s, nb_taps, tsrc);
	400	+ }
	401	+
	402	+ if (s->ir_link) {
	403	+ float gain = +INFINITY;
	404	+
	405	+ for (int ch = 0; ch < s->nb_channels; ch++)
	406	+ gain = fminf(gain, s->ch_gain[ch]);
	407	+
	408	+ for (int ch = 0; ch < s->nb_channels; ch++)
	409	+ s->ch_gain[ch] = gain;
	410	+ }
	411	+
396	412	for (int ch = 0; ch < s->nb_channels; ch++) {
397	413	const float tsrc = (const float )s->ir[selir]->extended_data[!s->one2many * ch];
398	414	float time = (float )s->norm_ir[selir]->extended_data[ch];
@@ -401,7 +417,7 @@ skip: · 2 unmodified lines
401	417	for (int i = FFMAX(1, s->length * nb_taps); i < nb_taps; i++)
402	418	time[i] = 0;
403	419
404		− get_power_float(ctx, s, nb_taps, ch, time);
	420	+ ir_scale_float(ctx, s, nb_taps, ch, time, s->ch_gain[ch]);
405	421
406	422	for (int n = 0; n < s->nb_segments[selir]; n++) {
407	423	AudioFIRSegment *seg = &s->seg[selir][n];
@@ -417,6 +433,22 @@ skip: · 9 unmodified lines
417	433	}
418	434	break;
419	435	case AV_SAMPLE_FMT_DBLP:
	436	+ for (int ch = 0; ch < s->nb_channels; ch++) {
	437	+ const double tsrc = (const double )s->ir[selir]->extended_data[!s->one2many * ch];
	438	+
	439	+ s->ch_gain[ch] = ir_gain_double(ctx, s, nb_taps, tsrc);
	440	+ }
	441	+
	442	+ if (s->ir_link) {
	443	+ double gain = +INFINITY;
	444	+
	445	+ for (int ch = 0; ch < s->nb_channels; ch++)
	446	+ gain = fmin(gain, s->ch_gain[ch]);
	447	+
	448	+ for (int ch = 0; ch < s->nb_channels; ch++)
	449	+ s->ch_gain[ch] = gain;
	450	+ }
	451	+
420	452	for (int ch = 0; ch < s->nb_channels; ch++) {
421	453	const double tsrc = (const double )s->ir[selir]->extended_data[!s->one2many * ch];
422	454	double time = (double )s->norm_ir[selir]->extended_data[ch];
@@ -425,7 +457,8 @@ skip: · 2 unmodified lines
425	457	for (int i = FFMAX(1, s->length * nb_taps); i < nb_taps; i++)
426	458	time[i] = 0;
427	459
428		− get_power_double(ctx, s, nb_taps, ch, time);
	460	+ ir_scale_double(ctx, s, nb_taps, ch, time, s->ch_gain[ch]);
	461	+
429	462	for (int n = 0; n < s->nb_segments[selir]; n++) {
430	463	AudioFIRSegment *seg = &s->seg[selir][n];
431	464
@@ -627,8 +660,9 @@ FF_ENABLE_DEPRECATION_WARNINGS · 195 unmodified lines
627	660
628	661	s->format = outlink->format;
629	662	s->nb_channels = outlink->ch_layout.nb_channels;
	663	+ s->ch_gain = av_calloc(ctx->inputs[0]->ch_layout.nb_channels, sizeof(*s->ch_gain));
630	664	s->loading = av_calloc(ctx->inputs[0]->ch_layout.nb_channels, sizeof(*s->loading));
631		− if (!s->loading)
	665	+ if (!s->loading \|\| !s->ch_gain)
632	666	return AVERROR(ENOMEM);
633	667
634	668	s->fadein[0] = ff_get_audio_buffer(outlink, s->min_part_size);
@@ -674,6 +708,7 @@ static av_cold void uninit(AVFilterContext *ctx) · 39 unmodified lines
674	708	AudioFIRContext *s = ctx->priv;
675	709
676	710	av_freep(&s->fdsp);
	711	+ av_freep(&s->ch_gain);
677	712	av_freep(&s->loading);
678	713
679	714	for (int i = 0; i < s->nb_irs; i++) {
@@ -812,13 +847,15 @@ static const AVOption afir_options[] = { · 132 unmodified lines
812	847	{ "dry", "set dry gain", OFFSET(dry_gain), AV_OPT_TYPE_FLOAT, {.dbl=1}, 0, 10, AFR },
813	848	{ "wet", "set wet gain", OFFSET(wet_gain), AV_OPT_TYPE_FLOAT, {.dbl=1}, 0, 10, AFR },
814	849	{ "length", "set IR length", OFFSET(length), AV_OPT_TYPE_FLOAT, {.dbl=1}, 0, 1, AF },
815		− { "gtype", "set IR auto gain type",OFFSET(gtype), AV_OPT_TYPE_INT, {.i64=0}, -1, 4, AF, "gtype" },
816		− { "none", "without auto gain", 0, AV_OPT_TYPE_CONST, {.i64=-1}, 0, 0, AF, "gtype" },
817		− { "peak", "peak gain", 0, AV_OPT_TYPE_CONST, {.i64=0}, 0, 0, AF, "gtype" },
818		− { "dc", "DC gain", 0, AV_OPT_TYPE_CONST, {.i64=1}, 0, 0, AF, "gtype" },
819		− { "gn", "gain to noise", 0, AV_OPT_TYPE_CONST, {.i64=2}, 0, 0, AF, "gtype" },
820		− { "ac", "AC gain", 0, AV_OPT_TYPE_CONST, {.i64=3}, 0, 0, AF, "gtype" },
821		− { "rms", "RMS gain", 0, AV_OPT_TYPE_CONST, {.i64=4}, 0, 0, AF, "gtype" },
	850	+ { "gtype", "set IR auto gain type",OFFSET(gtype), AV_OPT_TYPE_INT, {.i64=0}, -1, 4, AF\|AV_OPT_FLAG_DEPRECATED, "gtype" },
	851	+ { "none", "without auto gain", 0, AV_OPT_TYPE_CONST, {.i64=-1}, 0, 0, AF\|AV_OPT_FLAG_DEPRECATED, "gtype" },
	852	+ { "peak", "peak gain", 0, AV_OPT_TYPE_CONST, {.i64=0}, 0, 0, AF\|AV_OPT_FLAG_DEPRECATED, "gtype" },
	853	+ { "dc", "DC gain", 0, AV_OPT_TYPE_CONST, {.i64=1}, 0, 0, AF\|AV_OPT_FLAG_DEPRECATED, "gtype" },
	854	+ { "gn", "gain to noise", 0, AV_OPT_TYPE_CONST, {.i64=2}, 0, 0, AF\|AV_OPT_FLAG_DEPRECATED, "gtype" },
	855	+ { "ac", "AC gain", 0, AV_OPT_TYPE_CONST, {.i64=3}, 0, 0, AF\|AV_OPT_FLAG_DEPRECATED, "gtype" },
	856	+ { "rms", "RMS gain", 0, AV_OPT_TYPE_CONST, {.i64=4}, 0, 0, AF\|AV_OPT_FLAG_DEPRECATED, "gtype" },
	857	+ { "irnorm", "set IR norm", OFFSET(ir_norm), AV_OPT_TYPE_FLOAT, {.dbl=1}, -1, 2, AF },
	858	+ { "irlink", "set IR link", OFFSET(ir_link), AV_OPT_TYPE_BOOL, {.i64=1}, 0, 1, AF },
822	859	{ "irgain", "set IR gain", OFFSET(ir_gain), AV_OPT_TYPE_FLOAT, {.dbl=1}, 0, 1, AF },
823	860	{ "irfmt", "set IR format", OFFSET(ir_format), AV_OPT_TYPE_INT, {.i64=1}, 0, 1, AF, "irfmt" },
824	861	{ "mono", "single channel", 0, AV_OPT_TYPE_CONST, {.i64=0}, 0, 0, AF, "irfmt" },

libavfilter/af_afir.h+3−0

@@ -63,6 +63,8 @@ typedef struct AudioFIRContext { · 62 unmodified lines
63	63	float dry_gain;
64	64	float length;
65	65	int gtype;
	66	+ float ir_norm;
	67	+ float ir_link;
66	68	float ir_gain;
67	69	int ir_format;
68	70	int ir_load;
@@ -87,6 +89,7 @@ typedef struct AudioFIRContext { · 18 unmodified lines
87	89	int nb_channels;
88	90	int one2many;
89	91	int *loading;
	92	+ double *ch_gain;
90	93
91	94	AudioFIRSegment seg[MAX_IR_STREAMS][1024];
92	95

libavfilter/afir_template.c+27−86

@@ -29,6 +29,8 @@ · 28 unmodified lines
29	29	#undef HYPOT
30	30	#undef SAMPLE_FORMAT
31	31	#undef TX_TYPE
	32	+#undef FABS
	33	+#undef POW
32	34	#if DEPTH == 32
33	35	#define SAMPLE_FORMAT float
34	36	#define SQRT sqrtf
@@ -36,6 +38,8 @@ · 1 unmodified line
36	38	#define ctype AVComplexFloat
37	39	#define ftype float
38	40	#define TX_TYPE AV_TX_FLOAT_RDFT
	41	+#define FABS fabsf
	42	+#define POW powf
39	43	#else
40	44	#define SAMPLE_FORMAT double
41	45	#define SQRT sqrt
@@ -43,6 +47,8 @@ · 1 unmodified line
43	47	#define ctype AVComplexDouble
44	48	#define ftype double
45	49	#define TX_TYPE AV_TX_DOUBLE_RDFT
	50	+#define FABS fabs
	51	+#define POW pow
46	52	#endif
47	53
48	54	#define fn3(a,b) a##_##b
@@ -139,95 +145,32 @@ end: · 90 unmodified lines
139	145	av_free(mag);
140	146	}
141	147
142		−static int fn(get_power)(AVFilterContext ctx, AudioFIRContext s,
143		− int cur_nb_taps, int ch,
144		− ftype *time)
	148	+static ftype fn(ir_gain)(AVFilterContext ctx, AudioFIRContext s,
	149	+ int cur_nb_taps, const ftype *time)
145	150	{
146		− ftype ch_gain = 1;
	151	+ ftype ch_gain, sum = 0;
147	152
148		− switch (s->gtype) {
149		− case -1:
	153	+ if (s->ir_norm < 0.f) {
150	154	ch_gain = 1;
151		− break;
152		− case 0:
153		− {
154		− ftype sum = 0;
155		−
156		− for (int i = 0; i < cur_nb_taps; i++)
157		− sum += FFABS(time[i]);
158		− ch_gain = 1. / sum;
159		− }
160		− break;
161		− case 1:
162		− {
163		− ftype sum = 0;
164		−
165		− for (int i = 0; i < cur_nb_taps; i++)
166		− sum += time[i];
167		− ch_gain = 1. / sum;
168		− }
169		− break;
170		− case 2:
171		− {
172		− ftype sum = 0;
173		−
174		− for (int i = 0; i < cur_nb_taps; i++)
175		− sum += time[i] * time[i];
176		− ch_gain = 1. / SQRT(sum);
177		− }
178		− break;
179		− case 3:
180		− case 4:
181		− {
182		− ftype inc, outc, scale, power;
183		− AVTXContext *tx;
184		− av_tx_fn tx_fn;
185		− int ret, size;
186		−
187		− size = 1 << av_ceil_log2_c(cur_nb_taps);
188		− inc = av_calloc(size + 2, sizeof(SAMPLE_FORMAT));
189		− outc = av_calloc(size + 2, sizeof(SAMPLE_FORMAT));
190		− if (!inc \|\| !outc) {
191		− av_free(outc);
192		− av_free(inc);
193		− break;
194		− }
195		−
196		− scale = 1.;
197		− ret = av_tx_init(&tx, &tx_fn, TX_TYPE, 0, size, &scale, 0);
198		− if (ret < 0) {
199		− av_free(outc);
200		− av_free(inc);
201		− break;
202		− }
203		−
204		− {
205		− memcpy(inc, time, cur_nb_taps * sizeof(SAMPLE_FORMAT));
206		− tx_fn(tx, outc, inc, sizeof(SAMPLE_FORMAT));
207		−
208		− power = 0;
209		− if (s->gtype == 3) {
210		− for (int i = 0; i < size / 2 + 1; i++)
211		− power = FFMAX(power, HYPOT(outc[i * 2], outc[i * 2 + 1]));
212		− } else {
213		− ftype sum = 0;
214		− for (int i = 0; i < size / 2 + 1; i++)
215		− sum += HYPOT(outc[i * 2], outc[i * 2 + 1]);
216		− power = SQRT(sum / (size / 2 + 1));
217		− }
218		−
219		− ch_gain = 1. / power;
220		− }
221		−
222		− av_tx_uninit(&tx);
223		− av_free(outc);
224		− av_free(inc);
225		− }
226		− break;
227		− default:
228		− return AVERROR_BUG;
	155	+ } else if (s->ir_norm == 0.f) {
	156	+ for (int i = 0; i < cur_nb_taps; i++)
	157	+ sum += time[i];
	158	+ ch_gain = 1. / sum;
	159	+ } else {
	160	+ ftype ir_norm = s->ir_norm;
	161	+
	162	+ for (int i = 0; i < cur_nb_taps; i++)
	163	+ sum += POW(FABS(time[i]), ir_norm);
	164	+ ch_gain = 1. / POW(sum, 1. / ir_norm);
229	165	}
230	166
	167	+ return ch_gain;
	168	+}
	169	+
	170	+static void fn(ir_scale)(AVFilterContext ctx, AudioFIRContext s,
	171	+ int cur_nb_taps, int ch,
	172	+ ftype *time, ftype ch_gain)
	173	+{
231	174	if (ch_gain != 1. \|\| s->ir_gain != 1.) {
232	175	ftype gain = ch_gain * s->ir_gain;
233	176
@@ -238,8 +181,6 @@ static int fn(get_power)(AVFilterContext ctx, AudioFIRContext s, · 4 unmodified lines
238	181	s->fdsp->vector_dmul_scalar(time, time, gain, FFALIGN(cur_nb_taps, 8));
239	182	#endif
240	183	}
241		−
242		− return 0;
243	184	}
244	185
245	186	static void fn(convert_channel)(AVFilterContext ctx, AudioFIRContext s, int ch,

avfilter/af_afir: add irnorm and irlink options

Keyboard shortcuts

General

Issues and pull requests

Issues, pull requests and Inbox

Repository code