2 * Copyright (c) 2019 Paul B Mahol
4 * This file is part of FFmpeg.
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.
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.
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
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"
41 typedef struct AudioNLMSContext {
58 AVFloatDSPContext *fdsp;
61 #define OFFSET(x) offsetof(AudioNLMSContext, x)
62 #define A AV_OPT_FLAG_AUDIO_PARAM|AV_OPT_FLAG_FILTERING_PARAM
64 static const AVOption anlms_options[] = {
65 { "order", "set the filter order", OFFSET(order), AV_OPT_TYPE_INT, {.i64=256}, 1, INT16_MAX, A },
66 { "mu", "set the filter mu", OFFSET(mu), AV_OPT_TYPE_FLOAT, {.dbl=0.75}, 0, 2, A },
67 { "eps", "set the filter eps", OFFSET(eps), AV_OPT_TYPE_FLOAT, {.dbl=1}, 0, 1, A },
68 { "leakage", "set the filter leakage", OFFSET(leakage), AV_OPT_TYPE_FLOAT, {.dbl=0}, 0, 1, A },
69 { "out_mode", "set output mode", OFFSET(output_mode), AV_OPT_TYPE_INT, {.i64=OUT_MODE}, 0, NB_OMODES-1, A, "mode" },
70 { "i", "input", 0, AV_OPT_TYPE_CONST, {.i64=IN_MODE}, 0, 0, A, "mode" },
71 { "d", "desired", 0, AV_OPT_TYPE_CONST, {.i64=DESIRED_MODE}, 0, 0, A, "mode" },
72 { "o", "output", 0, AV_OPT_TYPE_CONST, {.i64=OUT_MODE}, 0, 0, A, "mode" },
73 { "n", "noise", 0, AV_OPT_TYPE_CONST, {.i64=NOISE_MODE}, 0, 0, A, "mode" },
77 AVFILTER_DEFINE_CLASS(anlms);
79 static int query_formats(AVFilterContext *ctx)
81 AVFilterFormats *formats;
82 AVFilterChannelLayouts *layouts;
83 static const enum AVSampleFormat sample_fmts[] = {
89 layouts = ff_all_channel_counts();
91 return AVERROR(ENOMEM);
92 ret = ff_set_common_channel_layouts(ctx, layouts);
96 formats = ff_make_format_list(sample_fmts);
98 return AVERROR(ENOMEM);
99 ret = ff_set_common_formats(ctx, formats);
103 formats = ff_all_samplerates();
105 return AVERROR(ENOMEM);
106 return ff_set_common_samplerates(ctx, formats);
109 static float fir_sample(AudioNLMSContext *s, float sample, float *delay,
110 float *coeffs, float *tmp, int *offset)
112 const int order = s->order;
115 delay[*offset] = sample;
117 memcpy(tmp, coeffs + order - *offset, order * sizeof(float));
119 output = s->fdsp->scalarproduct_float(delay, tmp, s->kernel_size);
127 static float process_sample(AudioNLMSContext *s, float input, float desired,
128 float *delay, float *coeffs, float *tmp, int *offsetp)
130 const int order = s->order;
131 const float leakage = s->leakage;
132 const float mu = s->mu;
133 const float a = 1.f - leakage * mu;
134 float sum, output, e, norm, b;
135 int offset = *offsetp;
137 delay[offset + order] = input;
139 output = fir_sample(s, input, delay, coeffs, tmp, offsetp);
140 e = desired - output;
142 sum = s->fdsp->scalarproduct_float(delay, delay, s->kernel_size);
147 memcpy(tmp, delay + offset, order * sizeof(float));
149 s->fdsp->vector_fmul_scalar(coeffs, coeffs, a, s->kernel_size);
151 s->fdsp->vector_fmac_scalar(coeffs, tmp, b, s->kernel_size);
153 memcpy(coeffs + order, coeffs, order * sizeof(float));
155 switch (s->output_mode) {
156 case IN_MODE: output = input; break;
157 case DESIRED_MODE: output = desired; break;
158 case OUT_MODE: /*output = output;*/ break;
159 case NOISE_MODE: output = desired - output; break;
164 static int process_channels(AVFilterContext *ctx, void *arg, int jobnr, int nb_jobs)
166 AudioNLMSContext *s = ctx->priv;
168 const int start = (out->channels * jobnr) / nb_jobs;
169 const int end = (out->channels * (jobnr+1)) / nb_jobs;
171 for (int c = start; c < end; c++) {
172 const float *input = (const float *)s->frame[0]->extended_data[c];
173 const float *desired = (const float *)s->frame[1]->extended_data[c];
174 float *delay = (float *)s->delay->extended_data[c];
175 float *coeffs = (float *)s->coeffs->extended_data[c];
176 float *tmp = (float *)s->tmp->extended_data[c];
177 int *offset = (int *)s->offset->extended_data[c];
178 float *output = (float *)out->extended_data[c];
180 for (int n = 0; n < out->nb_samples; n++)
181 output[n] = process_sample(s, input[n], desired[n], delay, coeffs, tmp, offset);
187 static int activate(AVFilterContext *ctx)
189 AudioNLMSContext *s = ctx->priv;
194 FF_FILTER_FORWARD_STATUS_BACK_ALL(ctx->outputs[0], ctx);
196 nb_samples = FFMIN(ff_inlink_queued_samples(ctx->inputs[0]),
197 ff_inlink_queued_samples(ctx->inputs[1]));
198 for (i = 0; i < ctx->nb_inputs && nb_samples > 0; i++) {
202 if (ff_inlink_check_available_samples(ctx->inputs[i], nb_samples) > 0) {
203 ret = ff_inlink_consume_samples(ctx->inputs[i], nb_samples, nb_samples, &s->frame[i]);
209 if (s->frame[0] && s->frame[1]) {
212 out = ff_get_audio_buffer(ctx->outputs[0], s->frame[0]->nb_samples);
214 av_frame_free(&s->frame[0]);
215 av_frame_free(&s->frame[1]);
216 return AVERROR(ENOMEM);
219 ctx->internal->execute(ctx, process_channels, out, NULL, FFMIN(ctx->outputs[0]->channels,
220 ff_filter_get_nb_threads(ctx)));
222 out->pts = s->frame[0]->pts;
224 av_frame_free(&s->frame[0]);
225 av_frame_free(&s->frame[1]);
227 ret = ff_filter_frame(ctx->outputs[0], out);
233 for (i = 0; i < 2; i++) {
234 if (ff_inlink_acknowledge_status(ctx->inputs[i], &status, &pts)) {
235 ff_outlink_set_status(ctx->outputs[0], status, pts);
241 if (ff_outlink_frame_wanted(ctx->outputs[0])) {
242 for (i = 0; i < 2; i++) {
243 if (ff_inlink_queued_samples(ctx->inputs[i]) > 0)
245 ff_inlink_request_frame(ctx->inputs[i]);
252 static int config_output(AVFilterLink *outlink)
254 AVFilterContext *ctx = outlink->src;
255 AudioNLMSContext *s = ctx->priv;
257 s->kernel_size = FFALIGN(s->order, 16);
260 s->offset = ff_get_audio_buffer(outlink, 1);
262 s->delay = ff_get_audio_buffer(outlink, 2 * s->kernel_size);
264 s->coeffs = ff_get_audio_buffer(outlink, 2 * s->kernel_size);
266 s->tmp = ff_get_audio_buffer(outlink, s->kernel_size);
267 if (!s->delay || !s->coeffs || !s->offset || !s->tmp)
268 return AVERROR(ENOMEM);
273 static av_cold int init(AVFilterContext *ctx)
275 AudioNLMSContext *s = ctx->priv;
277 s->fdsp = avpriv_float_dsp_alloc(0);
279 return AVERROR(ENOMEM);
284 static int process_command(AVFilterContext *ctx, const char *cmd, const char *args,
285 char *res, int res_len, int flags)
287 AudioNLMSContext *s = ctx->priv;
290 if ( !strcmp(cmd, "mu") || !strcmp(cmd, "eps")
291 || !strcmp(cmd, "leakage") || !strcmp(cmd, "out_mode")) {
292 ret = av_opt_set(s, cmd, args, 0);
294 ret = AVERROR(ENOSYS);
300 static av_cold void uninit(AVFilterContext *ctx)
302 AudioNLMSContext *s = ctx->priv;
305 av_frame_free(&s->delay);
306 av_frame_free(&s->coeffs);
307 av_frame_free(&s->offset);
308 av_frame_free(&s->tmp);
311 static const AVFilterPad inputs[] = {
314 .type = AVMEDIA_TYPE_AUDIO,
318 .type = AVMEDIA_TYPE_AUDIO,
323 static const AVFilterPad outputs[] = {
326 .type = AVMEDIA_TYPE_AUDIO,
327 .config_props = config_output,
332 AVFilter ff_af_anlms = {
334 .description = NULL_IF_CONFIG_SMALL("Apply Normalized Least-Mean-Squares algorithm to first audio stream."),
335 .priv_size = sizeof(AudioNLMSContext),
336 .priv_class = &anlms_class,
339 .activate = activate,
340 .query_formats = query_formats,
343 .flags = AVFILTER_FLAG_SLICE_THREADS,
344 .process_command = process_command,