1 |
|
|
/* |
2 |
|
|
* Copyright (c) 2019 Paul B Mahol |
3 |
|
|
* |
4 |
|
|
* This file is part of FFmpeg. |
5 |
|
|
* |
6 |
|
|
* FFmpeg 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 |
|
|
* FFmpeg 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 FFmpeg; if not, write to the Free Software |
18 |
|
|
* Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA |
19 |
|
|
*/ |
20 |
|
|
|
21 |
|
|
#include "libavutil/avassert.h" |
22 |
|
|
#include "libavutil/channel_layout.h" |
23 |
|
|
#include "libavutil/common.h" |
24 |
|
|
#include "libavutil/float_dsp.h" |
25 |
|
|
#include "libavutil/opt.h" |
26 |
|
|
|
27 |
|
|
#include "audio.h" |
28 |
|
|
#include "avfilter.h" |
29 |
|
|
#include "formats.h" |
30 |
|
|
#include "filters.h" |
31 |
|
|
#include "internal.h" |
32 |
|
|
|
33 |
|
|
enum OutModes { |
34 |
|
|
IN_MODE, |
35 |
|
|
DESIRED_MODE, |
36 |
|
|
OUT_MODE, |
37 |
|
|
NOISE_MODE, |
38 |
|
|
NB_OMODES |
39 |
|
|
}; |
40 |
|
|
|
41 |
|
|
typedef struct AudioNLMSContext { |
42 |
|
|
const AVClass *class; |
43 |
|
|
|
44 |
|
|
int order; |
45 |
|
|
float mu; |
46 |
|
|
float eps; |
47 |
|
|
float leakage; |
48 |
|
|
int output_mode; |
49 |
|
|
|
50 |
|
|
int kernel_size; |
51 |
|
|
AVFrame *offset; |
52 |
|
|
AVFrame *delay; |
53 |
|
|
AVFrame *coeffs; |
54 |
|
|
AVFrame *tmp; |
55 |
|
|
|
56 |
|
|
AVFrame *frame[2]; |
57 |
|
|
|
58 |
|
|
AVFloatDSPContext *fdsp; |
59 |
|
|
} AudioNLMSContext; |
60 |
|
|
|
61 |
|
|
#define OFFSET(x) offsetof(AudioNLMSContext, x) |
62 |
|
|
#define A AV_OPT_FLAG_AUDIO_PARAM|AV_OPT_FLAG_FILTERING_PARAM |
63 |
|
|
#define AT AV_OPT_FLAG_AUDIO_PARAM|AV_OPT_FLAG_FILTERING_PARAM|AV_OPT_FLAG_RUNTIME_PARAM |
64 |
|
|
|
65 |
|
|
static const AVOption anlms_options[] = { |
66 |
|
|
{ "order", "set the filter order", OFFSET(order), AV_OPT_TYPE_INT, {.i64=256}, 1, INT16_MAX, A }, |
67 |
|
|
{ "mu", "set the filter mu", OFFSET(mu), AV_OPT_TYPE_FLOAT, {.dbl=0.75}, 0, 2, AT }, |
68 |
|
|
{ "eps", "set the filter eps", OFFSET(eps), AV_OPT_TYPE_FLOAT, {.dbl=1}, 0, 1, AT }, |
69 |
|
|
{ "leakage", "set the filter leakage", OFFSET(leakage), AV_OPT_TYPE_FLOAT, {.dbl=0}, 0, 1, AT }, |
70 |
|
|
{ "out_mode", "set output mode", OFFSET(output_mode), AV_OPT_TYPE_INT, {.i64=OUT_MODE}, 0, NB_OMODES-1, AT, "mode" }, |
71 |
|
|
{ "i", "input", 0, AV_OPT_TYPE_CONST, {.i64=IN_MODE}, 0, 0, AT, "mode" }, |
72 |
|
|
{ "d", "desired", 0, AV_OPT_TYPE_CONST, {.i64=DESIRED_MODE}, 0, 0, AT, "mode" }, |
73 |
|
|
{ "o", "output", 0, AV_OPT_TYPE_CONST, {.i64=OUT_MODE}, 0, 0, AT, "mode" }, |
74 |
|
|
{ "n", "noise", 0, AV_OPT_TYPE_CONST, {.i64=NOISE_MODE}, 0, 0, AT, "mode" }, |
75 |
|
|
{ NULL } |
76 |
|
|
}; |
77 |
|
|
|
78 |
|
|
AVFILTER_DEFINE_CLASS(anlms); |
79 |
|
|
|
80 |
|
|
static int query_formats(AVFilterContext *ctx) |
81 |
|
|
{ |
82 |
|
|
AVFilterFormats *formats; |
83 |
|
|
AVFilterChannelLayouts *layouts; |
84 |
|
|
static const enum AVSampleFormat sample_fmts[] = { |
85 |
|
|
AV_SAMPLE_FMT_FLTP, |
86 |
|
|
AV_SAMPLE_FMT_NONE |
87 |
|
|
}; |
88 |
|
|
int ret; |
89 |
|
|
|
90 |
|
|
layouts = ff_all_channel_counts(); |
91 |
|
|
if (!layouts) |
92 |
|
|
return AVERROR(ENOMEM); |
93 |
|
|
ret = ff_set_common_channel_layouts(ctx, layouts); |
94 |
|
|
if (ret < 0) |
95 |
|
|
return ret; |
96 |
|
|
|
97 |
|
|
formats = ff_make_format_list(sample_fmts); |
98 |
|
|
if (!formats) |
99 |
|
|
return AVERROR(ENOMEM); |
100 |
|
|
ret = ff_set_common_formats(ctx, formats); |
101 |
|
|
if (ret < 0) |
102 |
|
|
return ret; |
103 |
|
|
|
104 |
|
|
formats = ff_all_samplerates(); |
105 |
|
|
if (!formats) |
106 |
|
|
return AVERROR(ENOMEM); |
107 |
|
|
return ff_set_common_samplerates(ctx, formats); |
108 |
|
|
} |
109 |
|
|
|
110 |
|
|
static float fir_sample(AudioNLMSContext *s, float sample, float *delay, |
111 |
|
|
float *coeffs, float *tmp, int *offset) |
112 |
|
|
{ |
113 |
|
|
const int order = s->order; |
114 |
|
|
float output; |
115 |
|
|
|
116 |
|
|
delay[*offset] = sample; |
117 |
|
|
|
118 |
|
|
memcpy(tmp, coeffs + order - *offset, order * sizeof(float)); |
119 |
|
|
|
120 |
|
|
output = s->fdsp->scalarproduct_float(delay, tmp, s->kernel_size); |
121 |
|
|
|
122 |
|
|
if (--(*offset) < 0) |
123 |
|
|
*offset = order - 1; |
124 |
|
|
|
125 |
|
|
return output; |
126 |
|
|
} |
127 |
|
|
|
128 |
|
|
static float process_sample(AudioNLMSContext *s, float input, float desired, |
129 |
|
|
float *delay, float *coeffs, float *tmp, int *offsetp) |
130 |
|
|
{ |
131 |
|
|
const int order = s->order; |
132 |
|
|
const float leakage = s->leakage; |
133 |
|
|
const float mu = s->mu; |
134 |
|
|
const float a = 1.f - leakage * mu; |
135 |
|
|
float sum, output, e, norm, b; |
136 |
|
|
int offset = *offsetp; |
137 |
|
|
|
138 |
|
|
delay[offset + order] = input; |
139 |
|
|
|
140 |
|
|
output = fir_sample(s, input, delay, coeffs, tmp, offsetp); |
141 |
|
|
e = desired - output; |
142 |
|
|
|
143 |
|
|
sum = s->fdsp->scalarproduct_float(delay, delay, s->kernel_size); |
144 |
|
|
|
145 |
|
|
norm = s->eps + sum; |
146 |
|
|
b = mu * e / norm; |
147 |
|
|
|
148 |
|
|
memcpy(tmp, delay + offset, order * sizeof(float)); |
149 |
|
|
|
150 |
|
|
s->fdsp->vector_fmul_scalar(coeffs, coeffs, a, s->kernel_size); |
151 |
|
|
|
152 |
|
|
s->fdsp->vector_fmac_scalar(coeffs, tmp, b, s->kernel_size); |
153 |
|
|
|
154 |
|
|
memcpy(coeffs + order, coeffs, order * sizeof(float)); |
155 |
|
|
|
156 |
|
|
switch (s->output_mode) { |
157 |
|
|
case IN_MODE: output = input; break; |
158 |
|
|
case DESIRED_MODE: output = desired; break; |
159 |
|
|
case OUT_MODE: /*output = output;*/ break; |
160 |
|
|
case NOISE_MODE: output = desired - output; break; |
161 |
|
|
} |
162 |
|
|
return output; |
163 |
|
|
} |
164 |
|
|
|
165 |
|
|
static int process_channels(AVFilterContext *ctx, void *arg, int jobnr, int nb_jobs) |
166 |
|
|
{ |
167 |
|
|
AudioNLMSContext *s = ctx->priv; |
168 |
|
|
AVFrame *out = arg; |
169 |
|
|
const int start = (out->channels * jobnr) / nb_jobs; |
170 |
|
|
const int end = (out->channels * (jobnr+1)) / nb_jobs; |
171 |
|
|
|
172 |
|
|
for (int c = start; c < end; c++) { |
173 |
|
|
const float *input = (const float *)s->frame[0]->extended_data[c]; |
174 |
|
|
const float *desired = (const float *)s->frame[1]->extended_data[c]; |
175 |
|
|
float *delay = (float *)s->delay->extended_data[c]; |
176 |
|
|
float *coeffs = (float *)s->coeffs->extended_data[c]; |
177 |
|
|
float *tmp = (float *)s->tmp->extended_data[c]; |
178 |
|
|
int *offset = (int *)s->offset->extended_data[c]; |
179 |
|
|
float *output = (float *)out->extended_data[c]; |
180 |
|
|
|
181 |
|
|
for (int n = 0; n < out->nb_samples; n++) |
182 |
|
|
output[n] = process_sample(s, input[n], desired[n], delay, coeffs, tmp, offset); |
183 |
|
|
} |
184 |
|
|
|
185 |
|
|
return 0; |
186 |
|
|
} |
187 |
|
|
|
188 |
|
|
static int activate(AVFilterContext *ctx) |
189 |
|
|
{ |
190 |
|
|
AudioNLMSContext *s = ctx->priv; |
191 |
|
|
int i, ret, status; |
192 |
|
|
int nb_samples; |
193 |
|
|
int64_t pts; |
194 |
|
|
|
195 |
|
|
FF_FILTER_FORWARD_STATUS_BACK_ALL(ctx->outputs[0], ctx); |
196 |
|
|
|
197 |
|
|
nb_samples = FFMIN(ff_inlink_queued_samples(ctx->inputs[0]), |
198 |
|
|
ff_inlink_queued_samples(ctx->inputs[1])); |
199 |
|
|
for (i = 0; i < ctx->nb_inputs && nb_samples > 0; i++) { |
200 |
|
|
if (s->frame[i]) |
201 |
|
|
continue; |
202 |
|
|
|
203 |
|
|
if (ff_inlink_check_available_samples(ctx->inputs[i], nb_samples) > 0) { |
204 |
|
|
ret = ff_inlink_consume_samples(ctx->inputs[i], nb_samples, nb_samples, &s->frame[i]); |
205 |
|
|
if (ret < 0) |
206 |
|
|
return ret; |
207 |
|
|
} |
208 |
|
|
} |
209 |
|
|
|
210 |
|
|
if (s->frame[0] && s->frame[1]) { |
211 |
|
|
AVFrame *out; |
212 |
|
|
|
213 |
|
|
out = ff_get_audio_buffer(ctx->outputs[0], s->frame[0]->nb_samples); |
214 |
|
|
if (!out) { |
215 |
|
|
av_frame_free(&s->frame[0]); |
216 |
|
|
av_frame_free(&s->frame[1]); |
217 |
|
|
return AVERROR(ENOMEM); |
218 |
|
|
} |
219 |
|
|
|
220 |
|
|
ctx->internal->execute(ctx, process_channels, out, NULL, FFMIN(ctx->outputs[0]->channels, |
221 |
|
|
ff_filter_get_nb_threads(ctx))); |
222 |
|
|
|
223 |
|
|
out->pts = s->frame[0]->pts; |
224 |
|
|
|
225 |
|
|
av_frame_free(&s->frame[0]); |
226 |
|
|
av_frame_free(&s->frame[1]); |
227 |
|
|
|
228 |
|
|
ret = ff_filter_frame(ctx->outputs[0], out); |
229 |
|
|
if (ret < 0) |
230 |
|
|
return ret; |
231 |
|
|
} |
232 |
|
|
|
233 |
|
|
if (!nb_samples) { |
234 |
|
|
for (i = 0; i < 2; i++) { |
235 |
|
|
if (ff_inlink_acknowledge_status(ctx->inputs[i], &status, &pts)) { |
236 |
|
|
ff_outlink_set_status(ctx->outputs[0], status, pts); |
237 |
|
|
return 0; |
238 |
|
|
} |
239 |
|
|
} |
240 |
|
|
} |
241 |
|
|
|
242 |
|
|
if (ff_outlink_frame_wanted(ctx->outputs[0])) { |
243 |
|
|
for (i = 0; i < 2; i++) { |
244 |
|
|
if (ff_inlink_queued_samples(ctx->inputs[i]) > 0) |
245 |
|
|
continue; |
246 |
|
|
ff_inlink_request_frame(ctx->inputs[i]); |
247 |
|
|
return 0; |
248 |
|
|
} |
249 |
|
|
} |
250 |
|
|
return 0; |
251 |
|
|
} |
252 |
|
|
|
253 |
|
|
static int config_output(AVFilterLink *outlink) |
254 |
|
|
{ |
255 |
|
|
AVFilterContext *ctx = outlink->src; |
256 |
|
|
AudioNLMSContext *s = ctx->priv; |
257 |
|
|
|
258 |
|
|
s->kernel_size = FFALIGN(s->order, 16); |
259 |
|
|
|
260 |
|
|
if (!s->offset) |
261 |
|
|
s->offset = ff_get_audio_buffer(outlink, 1); |
262 |
|
|
if (!s->delay) |
263 |
|
|
s->delay = ff_get_audio_buffer(outlink, 2 * s->kernel_size); |
264 |
|
|
if (!s->coeffs) |
265 |
|
|
s->coeffs = ff_get_audio_buffer(outlink, 2 * s->kernel_size); |
266 |
|
|
if (!s->tmp) |
267 |
|
|
s->tmp = ff_get_audio_buffer(outlink, s->kernel_size); |
268 |
|
|
if (!s->delay || !s->coeffs || !s->offset || !s->tmp) |
269 |
|
|
return AVERROR(ENOMEM); |
270 |
|
|
|
271 |
|
|
return 0; |
272 |
|
|
} |
273 |
|
|
|
274 |
|
|
static av_cold int init(AVFilterContext *ctx) |
275 |
|
|
{ |
276 |
|
|
AudioNLMSContext *s = ctx->priv; |
277 |
|
|
|
278 |
|
|
s->fdsp = avpriv_float_dsp_alloc(0); |
279 |
|
|
if (!s->fdsp) |
280 |
|
|
return AVERROR(ENOMEM); |
281 |
|
|
|
282 |
|
|
return 0; |
283 |
|
|
} |
284 |
|
|
|
285 |
|
|
static av_cold void uninit(AVFilterContext *ctx) |
286 |
|
|
{ |
287 |
|
|
AudioNLMSContext *s = ctx->priv; |
288 |
|
|
|
289 |
|
|
av_freep(&s->fdsp); |
290 |
|
|
av_frame_free(&s->delay); |
291 |
|
|
av_frame_free(&s->coeffs); |
292 |
|
|
av_frame_free(&s->offset); |
293 |
|
|
av_frame_free(&s->tmp); |
294 |
|
|
} |
295 |
|
|
|
296 |
|
|
static const AVFilterPad inputs[] = { |
297 |
|
|
{ |
298 |
|
|
.name = "input", |
299 |
|
|
.type = AVMEDIA_TYPE_AUDIO, |
300 |
|
|
}, |
301 |
|
|
{ |
302 |
|
|
.name = "desired", |
303 |
|
|
.type = AVMEDIA_TYPE_AUDIO, |
304 |
|
|
}, |
305 |
|
|
{ NULL } |
306 |
|
|
}; |
307 |
|
|
|
308 |
|
|
static const AVFilterPad outputs[] = { |
309 |
|
|
{ |
310 |
|
|
.name = "default", |
311 |
|
|
.type = AVMEDIA_TYPE_AUDIO, |
312 |
|
|
.config_props = config_output, |
313 |
|
|
}, |
314 |
|
|
{ NULL } |
315 |
|
|
}; |
316 |
|
|
|
317 |
|
|
AVFilter ff_af_anlms = { |
318 |
|
|
.name = "anlms", |
319 |
|
|
.description = NULL_IF_CONFIG_SMALL("Apply Normalized Least-Mean-Squares algorithm to first audio stream."), |
320 |
|
|
.priv_size = sizeof(AudioNLMSContext), |
321 |
|
|
.priv_class = &anlms_class, |
322 |
|
|
.init = init, |
323 |
|
|
.uninit = uninit, |
324 |
|
|
.activate = activate, |
325 |
|
|
.query_formats = query_formats, |
326 |
|
|
.inputs = inputs, |
327 |
|
|
.outputs = outputs, |
328 |
|
|
.flags = AVFILTER_FLAG_SLICE_THREADS, |
329 |
|
|
.process_command = ff_filter_process_command, |
330 |
|
|
}; |