avfilter: add arls filter

Changelog

@@ -6,6 +6,7 @@ version <next>:
 - Playdate video decoder and demuxer
 - Extend VAAPI support for libva-win32 on Windows
 - afireqsrc audio source filter
+- arls filter
 
 version 6.0:
 - Radiance HDR image support

doc/filters.texi

@@ -2975,6 +2975,44 @@ atrim=end=5,areverse
 @end example
 @end itemize
 
+@section arls
+Apply Recursive Least Squares algorithm to the first audio stream using the second audio stream.
+
+This adaptive filter is used to mimic a desired filter by recursively finding the filter coefficients that
+relate to producing the minimal weighted linear least squares cost function of the error signal (difference
+between the desired, 2nd input audio stream and the actual signal, the 1st input audio stream).
+
+A description of the accepted options follows.
+
+@table @option
+@item order
+Set the filter order.
+
+@item lambda
+Set the forgetting factor.
+
+@item delta
+Set the coefficient to initialize internal covariance matrix.
+
+@item out_mode
+Set the filter output samples. It accepts the following values:
+@table @option
+@item i
+Pass the 1st input.
+
+@item d
+Pass the 2nd input.
+
+@item o
+Pass filtered samples.
+
+@item n
+Pass difference between desired and filtered samples.
+
+Default value is @var{o}.
+@end table
+@end table
+
 @section arnndn
 
 Reduce noise from speech using Recurrent Neural Networks.

libavfilter/Makefile

@@ -88,6 +88,7 @@ OBJS-$(CONFIG_APULSATOR_FILTER)              += af_apulsator.o
 OBJS-$(CONFIG_AREALTIME_FILTER)              += f_realtime.o
 OBJS-$(CONFIG_ARESAMPLE_FILTER)              += af_aresample.o
 OBJS-$(CONFIG_AREVERSE_FILTER)               += f_reverse.o
+OBJS-$(CONFIG_ARLS_FILTER)                   += af_arls.o
 OBJS-$(CONFIG_ARNNDN_FILTER)                 += af_arnndn.o
 OBJS-$(CONFIG_ASDR_FILTER)                   += af_asdr.o
 OBJS-$(CONFIG_ASEGMENT_FILTER)               += f_segment.o

libavfilter/af_arls.c

@@ -0,0 +1,353 @@
+/*
+ * Copyright (c) 2023 Paul B Mahol
+ *
+ * This file is part of FFmpeg.
+ *
+ * FFmpeg is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU Lesser General Public
+ * License as published by the Free Software Foundation; either
+ * version 2.1 of the License, or (at your option) any later version.
+ *
+ * FFmpeg is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+ * Lesser General Public License for more details.
+ *
+ * You should have received a copy of the GNU Lesser General Public
+ * License along with FFmpeg; if not, write to the Free Software
+ * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
+ */
+
+#include "libavutil/common.h"
+#include "libavutil/float_dsp.h"
+#include "libavutil/opt.h"
+
+#include "audio.h"
+#include "avfilter.h"
+#include "formats.h"
+#include "filters.h"
+#include "internal.h"
+
+enum OutModes {
+    IN_MODE,
+    DESIRED_MODE,
+    OUT_MODE,
+    NOISE_MODE,
+    NB_OMODES
+};
+
+typedef struct AudioRLSContext {
+    const AVClass *class;
+
+    int order;
+    float lambda;
+    float delta;
+    int output_mode;
+
+    int kernel_size;
+    AVFrame *offset;
+    AVFrame *delay;
+    AVFrame *coeffs;
+    AVFrame *p, *dp;
+    AVFrame *gains;
+    AVFrame *u, *tmp;
+
+    AVFrame *frame[2];
+
+    AVFloatDSPContext *fdsp;
+} AudioRLSContext;
+
+#define OFFSET(x) offsetof(AudioRLSContext, x)
+#define A AV_OPT_FLAG_AUDIO_PARAM|AV_OPT_FLAG_FILTERING_PARAM
+#define AT AV_OPT_FLAG_AUDIO_PARAM|AV_OPT_FLAG_FILTERING_PARAM|AV_OPT_FLAG_RUNTIME_PARAM
+
+static const AVOption arls_options[] = {
+    { "order",    "set the filter order",  OFFSET(order),  AV_OPT_TYPE_INT,   {.i64=16}, 1, INT16_MAX, A },
+    { "lambda",   "set the filter lambda", OFFSET(lambda), AV_OPT_TYPE_FLOAT, {.dbl=1.f}, 0, 1, AT },
+    { "delta",    "set the filter delta",  OFFSET(delta),  AV_OPT_TYPE_FLOAT, {.dbl=2.f}, 0, INT16_MAX, A },
+    { "out_mode", "set output mode",       OFFSET(output_mode), AV_OPT_TYPE_INT, {.i64=OUT_MODE}, 0, NB_OMODES-1, AT, "mode" },
+    {  "i", "input",   0, AV_OPT_TYPE_CONST, {.i64=IN_MODE},      0, 0, AT, "mode" },
+    {  "d", "desired", 0, AV_OPT_TYPE_CONST, {.i64=DESIRED_MODE}, 0, 0, AT, "mode" },
+    {  "o", "output",  0, AV_OPT_TYPE_CONST, {.i64=OUT_MODE},     0, 0, AT, "mode" },
+    {  "n", "noise",   0, AV_OPT_TYPE_CONST, {.i64=NOISE_MODE},   0, 0, AT, "mode" },
+    { NULL }
+};
+
+AVFILTER_DEFINE_CLASS(arls);
+
+static float fir_sample(AudioRLSContext *s, float sample, float *delay,
+                        float *coeffs, float *tmp, int *offset)
+{
+    const int order = s->order;
+    float output;
+
+    delay[*offset] = sample;
+
+    memcpy(tmp, coeffs + order - *offset, order * sizeof(float));
+
+    output = s->fdsp->scalarproduct_float(delay, tmp, s->kernel_size);
+
+    if (--(*offset) < 0)
+        *offset = order - 1;
+
+    return output;
+}
+
+static float process_sample(AudioRLSContext *s, float input, float desired, int ch)
+{
+    float *coeffs = (float *)s->coeffs->extended_data[ch];
+    float *delay = (float *)s->delay->extended_data[ch];
+    float *gains = (float *)s->gains->extended_data[ch];
+    float *tmp = (float *)s->tmp->extended_data[ch];
+    float *u = (float *)s->u->extended_data[ch];
+    float *p = (float *)s->p->extended_data[ch];
+    float *dp = (float *)s->dp->extended_data[ch];
+    int *offsetp = (int *)s->offset->extended_data[ch];
+    const int kernel_size = s->kernel_size;
+    const int order = s->order;
+    const float lambda = s->lambda;
+    int offset = *offsetp;
+    float g = lambda;
+    float output, e;
+
+    delay[offset + order] = input;
+
+    output = fir_sample(s, input, delay, coeffs, tmp, offsetp);
+    e = desired - output;
+
+    for (int i = 0, pos = offset; i < order; i++, pos++) {
+        const int ikernel_size = i * kernel_size;
+
+        u[i] = 0.f;
+        for (int k = 0, pos = offset; k < order; k++, pos++)
+            u[i] += p[ikernel_size + k] * delay[pos];
+
+        g += u[i] * delay[pos];
+    }
+
+    g = 1.f / g;
+
+    for (int i = 0; i < order; i++) {
+        const int ikernel_size = i * kernel_size;
+
+        gains[i] = u[i] * g;
+        coeffs[i] = coeffs[order + i] = coeffs[i] + gains[i] * e;
+        tmp[i] = 0.f;
+        for (int k = 0, pos = offset; k < order; k++, pos++)
+            tmp[i] += p[ikernel_size + k] * delay[pos];
+    }
+
+    for (int i = 0; i < order; i++) {
+        const int ikernel_size = i * kernel_size;
+
+        for (int k = 0; k < order; k++)
+            dp[ikernel_size + k] = gains[i] * tmp[k];
+    }
+
+    for (int i = 0; i < order; i++) {
+        const int ikernel_size = i * kernel_size;
+
+        for (int k = 0; k < order; k++)
+            p[ikernel_size + k] = (p[ikernel_size + k] - (dp[ikernel_size + k] + dp[kernel_size * k + i]) * 0.5f) * lambda;
+    }
+
+    switch (s->output_mode) {
+    case IN_MODE:       output = input;         break;
+    case DESIRED_MODE:  output = desired;       break;
+    case OUT_MODE:   output = desired - output; break;
+    case NOISE_MODE: output = input - output;   break;
+    }
+    return output;
+}
+
+static int process_channels(AVFilterContext *ctx, void *arg, int jobnr, int nb_jobs)
+{
+    AudioRLSContext *s = ctx->priv;
+    AVFrame *out = arg;
+    const int start = (out->ch_layout.nb_channels * jobnr) / nb_jobs;
+    const int end = (out->ch_layout.nb_channels * (jobnr+1)) / nb_jobs;
+
+    for (int c = start; c < end; c++) {
+        const float *input = (const float *)s->frame[0]->extended_data[c];
+        const float *desired = (const float *)s->frame[1]->extended_data[c];
+        float *output = (float *)out->extended_data[c];
+
+        for (int n = 0; n < out->nb_samples; n++) {
+            output[n] = process_sample(s, input[n], desired[n], c);
+            if (ctx->is_disabled)
+                output[n] = input[n];
+        }
+    }
+
+    return 0;
+}
+
+static int activate(AVFilterContext *ctx)
+{
+    AudioRLSContext *s = ctx->priv;
+    int i, ret, status;
+    int nb_samples;
+    int64_t pts;
+
+    FF_FILTER_FORWARD_STATUS_BACK_ALL(ctx->outputs[0], ctx);
+
+    nb_samples = FFMIN(ff_inlink_queued_samples(ctx->inputs[0]),
+                       ff_inlink_queued_samples(ctx->inputs[1]));
+    for (i = 0; i < ctx->nb_inputs && nb_samples > 0; i++) {
+        if (s->frame[i])
+            continue;
+
+        if (ff_inlink_check_available_samples(ctx->inputs[i], nb_samples) > 0) {
+            ret = ff_inlink_consume_samples(ctx->inputs[i], nb_samples, nb_samples, &s->frame[i]);
+            if (ret < 0)
+                return ret;
+        }
+    }
+
+    if (s->frame[0] && s->frame[1]) {
+        AVFrame *out;
+
+        out = ff_get_audio_buffer(ctx->outputs[0], s->frame[0]->nb_samples);
+        if (!out) {
+            av_frame_free(&s->frame[0]);
+            av_frame_free(&s->frame[1]);
+            return AVERROR(ENOMEM);
+        }
+
+        ff_filter_execute(ctx, process_channels, out, NULL,
+                          FFMIN(ctx->outputs[0]->ch_layout.nb_channels, ff_filter_get_nb_threads(ctx)));
+
+        out->pts = s->frame[0]->pts;
+
+        av_frame_free(&s->frame[0]);
+        av_frame_free(&s->frame[1]);
+
+        ret = ff_filter_frame(ctx->outputs[0], out);
+        if (ret < 0)
+            return ret;
+    }
+
+    if (!nb_samples) {
+        for (i = 0; i < 2; i++) {
+            if (ff_inlink_acknowledge_status(ctx->inputs[i], &status, &pts)) {
+                ff_outlink_set_status(ctx->outputs[0], status, pts);
+                return 0;
+            }
+        }
+    }
+
+    if (ff_outlink_frame_wanted(ctx->outputs[0])) {
+        for (i = 0; i < 2; i++) {
+            if (ff_inlink_queued_samples(ctx->inputs[i]) > 0)
+                continue;
+            ff_inlink_request_frame(ctx->inputs[i]);
+            return 0;
+        }
+    }
+    return 0;
+}
+
+static int config_output(AVFilterLink *outlink)
+{
+    AVFilterContext *ctx = outlink->src;
+    AudioRLSContext *s = ctx->priv;
+
+    s->kernel_size = FFALIGN(s->order, 16);
+
+    if (!s->offset)
+        s->offset = ff_get_audio_buffer(outlink, 1);
+    if (!s->delay)
+        s->delay = ff_get_audio_buffer(outlink, 2 * s->kernel_size);
+    if (!s->coeffs)
+        s->coeffs = ff_get_audio_buffer(outlink, 2 * s->kernel_size);
+    if (!s->gains)
+        s->gains = ff_get_audio_buffer(outlink, s->kernel_size);
+    if (!s->p)
+        s->p = ff_get_audio_buffer(outlink, s->kernel_size * s->kernel_size);
+    if (!s->dp)
+        s->dp = ff_get_audio_buffer(outlink, s->kernel_size * s->kernel_size);
+    if (!s->u)
+        s->u = ff_get_audio_buffer(outlink, s->kernel_size);
+    if (!s->tmp)
+        s->tmp = ff_get_audio_buffer(outlink, s->kernel_size);
+
+    if (!s->delay || !s->coeffs || !s->p || !s->dp || !s->gains || !s->offset || !s->u || !s->tmp)
+        return AVERROR(ENOMEM);
+
+    for (int ch = 0; ch < s->offset->ch_layout.nb_channels; ch++) {
+        int *dst = (int *)s->offset->extended_data[ch];
+
+        for (int i = 0; i < s->kernel_size; i++)
+            dst[0] = s->kernel_size - 1;
+    }
+
+    for (int ch = 0; ch < s->p->ch_layout.nb_channels; ch++) {
+        float *dst = (float *)s->p->extended_data[ch];
+
+        for (int i = 0; i < s->kernel_size; i++)
+            dst[i * s->kernel_size + i] = s->delta;
+    }
+
+    return 0;
+}
+
+static av_cold int init(AVFilterContext *ctx)
+{
+    AudioRLSContext *s = ctx->priv;
+
+    s->fdsp = avpriv_float_dsp_alloc(0);
+    if (!s->fdsp)
+        return AVERROR(ENOMEM);
+
+    return 0;
+}
+
+static av_cold void uninit(AVFilterContext *ctx)
+{
+    AudioRLSContext *s = ctx->priv;
+
+    av_freep(&s->fdsp);
+    av_frame_free(&s->delay);
+    av_frame_free(&s->coeffs);
+    av_frame_free(&s->gains);
+    av_frame_free(&s->offset);
+    av_frame_free(&s->p);
+    av_frame_free(&s->dp);
+    av_frame_free(&s->u);
+    av_frame_free(&s->tmp);
+}
+
+static const AVFilterPad inputs[] = {
+    {
+        .name = "input",
+        .type = AVMEDIA_TYPE_AUDIO,
+    },
+    {
+        .name = "desired",
+        .type = AVMEDIA_TYPE_AUDIO,
+    },
+};
+
+static const AVFilterPad outputs[] = {
+    {
+        .name         = "default",
+        .type         = AVMEDIA_TYPE_AUDIO,
+        .config_props = config_output,
+    },
+};
+
+const AVFilter ff_af_arls = {
+    .name           = "arls",
+    .description    = NULL_IF_CONFIG_SMALL("Apply Recursive Least Squares algorithm to first audio stream."),
+    .priv_size      = sizeof(AudioRLSContext),
+    .priv_class     = &arls_class,
+    .init           = init,
+    .uninit         = uninit,
+    .activate       = activate,
+    FILTER_INPUTS(inputs),
+    FILTER_OUTPUTS(outputs),
+    FILTER_SINGLE_SAMPLEFMT(AV_SAMPLE_FMT_FLTP),
+    .flags          = AVFILTER_FLAG_SUPPORT_TIMELINE_INTERNAL |
+                      AVFILTER_FLAG_SLICE_THREADS,
+    .process_command = ff_filter_process_command,
+};

libavfilter/allfilters.c

@@ -75,6 +75,7 @@ extern const AVFilter ff_af_apulsator;
 extern const AVFilter ff_af_arealtime;
 extern const AVFilter ff_af_aresample;
 extern const AVFilter ff_af_areverse;
+extern const AVFilter ff_af_arls;
 extern const AVFilter ff_af_arnndn;
 extern const AVFilter ff_af_asdr;
 extern const AVFilter ff_af_asegment;

libavfilter/version.h

@@ -31,7 +31,7 @@
 
 #include "version_major.h"
 
-#define LIBAVFILTER_VERSION_MINOR   6
+#define LIBAVFILTER_VERSION_MINOR   7
 #define LIBAVFILTER_VERSION_MICRO 100