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
23 #include "libavutil/avassert.h"
24 #include "libavutil/audio_fifo.h"
25 #include "libavutil/opt.h"
30 #include "af_anlmdndsp.h"
33 #define WEIGHT_LUT_NBITS 20
34 #define WEIGHT_LUT_SIZE (1<<WEIGHT_LUT_NBITS)
36 #define SQR(x) ((x) * (x))
38 typedef struct AudioNLMeansContext {
45 float pdiff_lut_scale;
46 float weight_lut[WEIGHT_LUT_SIZE];
61 AudioNLMDNDSPContext dsp;
62 } AudioNLMeansContext;
64 #define OFFSET(x) offsetof(AudioNLMeansContext, x)
65 #define AF AV_OPT_FLAG_AUDIO_PARAM|AV_OPT_FLAG_FILTERING_PARAM
67 static const AVOption anlmdn_options[] = {
68 { "s", "set denoising strength", OFFSET(a), AV_OPT_TYPE_FLOAT, {.dbl=1}, 1, 9999, AF },
69 { "p", "set patch duration", OFFSET(pd), AV_OPT_TYPE_DURATION, {.i64=2000}, 1000, 100000, AF },
70 { "r", "set research duration", OFFSET(rd), AV_OPT_TYPE_DURATION, {.i64=6000}, 2000, 300000, AF },
74 AVFILTER_DEFINE_CLASS(anlmdn);
76 static int query_formats(AVFilterContext *ctx)
78 AVFilterFormats *formats = NULL;
79 AVFilterChannelLayouts *layouts = NULL;
80 static const enum AVSampleFormat sample_fmts[] = {
86 formats = ff_make_format_list(sample_fmts);
88 return AVERROR(ENOMEM);
89 ret = ff_set_common_formats(ctx, formats);
93 layouts = ff_all_channel_counts();
95 return AVERROR(ENOMEM);
97 ret = ff_set_common_channel_layouts(ctx, layouts);
101 formats = ff_all_samplerates();
102 return ff_set_common_samplerates(ctx, formats);
105 static float compute_distance_ssd_c(const float *f1, const float *f2, ptrdiff_t K)
109 for (int k = -K; k <= K; k++)
110 distance += SQR(f1[k] - f2[k]);
115 static void compute_cache_c(float *cache, const float *f,
116 ptrdiff_t S, ptrdiff_t K,
117 ptrdiff_t i, ptrdiff_t jj)
121 for (int j = jj; j < jj + S; j++, v++)
122 cache[v] += -SQR(f[i - K - 1] - f[j - K - 1]) + SQR(f[i + K] - f[j + K]);
125 void ff_anlmdn_init(AudioNLMDNDSPContext *dsp)
127 dsp->compute_distance_ssd = compute_distance_ssd_c;
128 dsp->compute_cache = compute_cache_c;
131 ff_anlmdn_init_x86(dsp);
134 static int config_output(AVFilterLink *outlink)
136 AVFilterContext *ctx = outlink->src;
137 AudioNLMeansContext *s = ctx->priv;
140 s->K = av_rescale(s->pd, outlink->sample_rate, AV_TIME_BASE);
141 s->S = av_rescale(s->rd, outlink->sample_rate, AV_TIME_BASE);
143 s->pts = AV_NOPTS_VALUE;
145 s->N = s->H + (s->K + s->S) * 2;
147 av_log(ctx, AV_LOG_DEBUG, "K:%d S:%d H:%d N:%d\n", s->K, s->S, s->H, s->N);
149 av_frame_free(&s->in);
150 av_frame_free(&s->cache);
151 s->in = ff_get_audio_buffer(outlink, s->N);
153 return AVERROR(ENOMEM);
155 s->cache = ff_get_audio_buffer(outlink, s->S * 2);
157 return AVERROR(ENOMEM);
159 s->fifo = av_audio_fifo_alloc(outlink->format, outlink->channels, s->N);
161 return AVERROR(ENOMEM);
163 ret = av_audio_fifo_write(s->fifo, (void **)s->in->extended_data, s->K + s->S);
167 s->pdiff_lut_scale = 1.f / MAX_DIFF * WEIGHT_LUT_SIZE;
168 for (int i = 0; i < WEIGHT_LUT_SIZE; i++) {
169 float w = -i / s->pdiff_lut_scale;
171 s->weight_lut[i] = expf(w);
174 ff_anlmdn_init(&s->dsp);
179 static int filter_channel(AVFilterContext *ctx, void *arg, int ch, int nb_jobs)
181 AudioNLMeansContext *s = ctx->priv;
185 const float *f = (const float *)(s->in->extended_data[ch]) + K;
186 float *cache = (float *)s->cache->extended_data[ch];
187 const float sw = 32768.f / s->a;
188 float *dst = (float *)out->extended_data[ch] + s->offset;
190 for (int i = S; i < s->H + S; i++) {
191 float P = 0.f, Q = 0.f;
195 for (int j = i - S; j <= i + S; j++) {
198 cache[v++] = s->dsp.compute_distance_ssd(f + i, f + j, K);
201 s->dsp.compute_cache(cache, f, S, K, i, i - S);
202 s->dsp.compute_cache(cache + S, f, S, K, i, i + 1);
205 for (int j = 0; j < 2 * S; j++) {
206 const float distance = cache[j];
207 unsigned weight_lut_idx;
210 av_assert2(distance >= 0.f);
214 weight_lut_idx = w * s->pdiff_lut_scale;
215 av_assert2(weight_lut_idx < WEIGHT_LUT_SIZE);
216 w = s->weight_lut[weight_lut_idx];
217 P += w * f[i - S + j + (j >= S)];
230 static int filter_frame(AVFilterLink *inlink, AVFrame *in)
232 AVFilterContext *ctx = inlink->dst;
233 AVFilterLink *outlink = ctx->outputs[0];
234 AudioNLMeansContext *s = ctx->priv;
236 int available, wanted, ret;
238 if (s->pts == AV_NOPTS_VALUE)
241 ret = av_audio_fifo_write(s->fifo, (void **)in->extended_data,
246 available = av_audio_fifo_size(s->fifo);
247 wanted = (available / s->H) * s->H;
249 if (wanted >= s->H && available >= s->N) {
250 out = ff_get_audio_buffer(outlink, wanted);
252 return AVERROR(ENOMEM);
255 while (available >= s->N) {
256 ret = av_audio_fifo_peek(s->fifo, (void **)s->in->extended_data, s->N);
260 ctx->internal->execute(ctx, filter_channel, out, NULL, inlink->channels);
262 av_audio_fifo_drain(s->fifo, s->H);
270 out->nb_samples = s->offset;
273 return ff_filter_frame(outlink, out);
279 static av_cold void uninit(AVFilterContext *ctx)
281 AudioNLMeansContext *s = ctx->priv;
283 av_audio_fifo_free(s->fifo);
284 av_frame_free(&s->in);
285 av_frame_free(&s->cache);
288 static const AVFilterPad inputs[] = {
291 .type = AVMEDIA_TYPE_AUDIO,
292 .filter_frame = filter_frame,
297 static const AVFilterPad outputs[] = {
300 .type = AVMEDIA_TYPE_AUDIO,
301 .config_props = config_output,
306 AVFilter ff_af_anlmdn = {
308 .description = NULL_IF_CONFIG_SMALL("Reduce broadband noise from stream using Non-Local Means."),
309 .query_formats = query_formats,
310 .priv_size = sizeof(AudioNLMeansContext),
311 .priv_class = &anlmdn_class,
315 .flags = AVFILTER_FLAG_SLICE_THREADS,