[FFmpeg-devel] avfilter: add anlms filter

Submitted by Paul B Mahol on Oct. 4, 2019, 8:04 a.m.

Details

Message ID 20191004080446.32102-1-onemda@gmail.com
State New
Headers show

Commit Message

Paul B Mahol Oct. 4, 2019, 8:04 a.m.
Signed-off-by: Paul B Mahol <onemda@gmail.com>
---
 doc/filters.texi         |  23 +++
 libavfilter/Makefile     |   1 +
 libavfilter/af_anlms.c   | 308 +++++++++++++++++++++++++++++++++++++++
 libavfilter/allfilters.c |   1 +
 4 files changed, 333 insertions(+)
 create mode 100644 libavfilter/af_anlms.c

Comments

Reino Wijnsma Oct. 5, 2019, 10:24 a.m.
On 2019-10-04T10:04:46+0200, Paul B Mahol <onemda@gmail.com> wrote:

> +Apply Normalized Least-Mean-Squares algorithm to first audio stream using second audio stream.

Apply Normalized Least-Mean-Squares algorithm to [the] first audio stream using [the] second audio stream.

> +This is adaptive filter used to mimic a desired filter by finding the filter coefficients that

This [-] adaptive filter [is] used to mimic a desired filter by finding the filter coefficients that

> +relate to producing the least mean square of the error signal (difference between the desired,
> +2nd input audio stream and the actual signal, 1st input audio stream).

2nd input audio stream and the actual signal, [the] 1st input audio stream).

-- Reino
Paul B Mahol Oct. 6, 2019, 12:57 p.m.
On 10/5/19, Reino Wijnsma <rwijnsma@xs4all.nl> wrote:
> On 2019-10-04T10:04:46+0200, Paul B Mahol <onemda@gmail.com> wrote:
>
>> +Apply Normalized Least-Mean-Squares algorithm to first audio stream using
>> second audio stream.
>
> Apply Normalized Least-Mean-Squares algorithm to [the] first audio stream
> using [the] second audio stream.
>
>> +This is adaptive filter used to mimic a desired filter by finding the
>> filter coefficients that
>
> This [-] adaptive filter [is] used to mimic a desired filter by finding the
> filter coefficients that
>
>> +relate to producing the least mean square of the error signal (difference
>> between the desired,
>> +2nd input audio stream and the actual signal, 1st input audio stream).
>
> 2nd input audio stream and the actual signal, [the] 1st input audio stream).
>

Will apply with these corrected.


> -- Reino
> _______________________________________________
> ffmpeg-devel mailing list
> ffmpeg-devel@ffmpeg.org
> https://ffmpeg.org/mailman/listinfo/ffmpeg-devel
>
> To unsubscribe, visit link above, or email
> ffmpeg-devel-request@ffmpeg.org with subject "unsubscribe".

Patch hide | download patch | download mbox

diff --git a/doc/filters.texi b/doc/filters.texi
index fbc3a404dd..49884fe16e 100644
--- a/doc/filters.texi
+++ b/doc/filters.texi
@@ -1814,6 +1814,29 @@  Change output mode.
 Syntax for the command is : "i", "o" or "n" string.
 @end table
 
+@section anlms
+Apply Normalized Least-Mean-Squares algorithm to first audio stream using second audio stream.
+
+This is adaptive filter used to mimic a desired filter by finding the filter coefficients that
+relate to producing the least mean square of the error signal (difference between the desired,
+2nd input audio stream and the actual signal, 1st input audio stream).
+
+A description of the accepted options follows.
+
+@table @option
+@item order
+Set filter order.
+
+@item mu
+Set filter mu.
+
+@item eps
+Set the filter eps.
+
+@item leakage
+Set the filter leakage.
+@end table
+
 @section anull
 
 Pass the audio source unchanged to the output.
diff --git a/libavfilter/Makefile b/libavfilter/Makefile
index 182fe9df4b..16bb8cd965 100644
--- a/libavfilter/Makefile
+++ b/libavfilter/Makefile
@@ -63,6 +63,7 @@  OBJS-$(CONFIG_AMIX_FILTER)                   += af_amix.o
 OBJS-$(CONFIG_AMULTIPLY_FILTER)              += af_amultiply.o
 OBJS-$(CONFIG_ANEQUALIZER_FILTER)            += af_anequalizer.o
 OBJS-$(CONFIG_ANLMDN_FILTER)                 += af_anlmdn.o
+OBJS-$(CONFIG_ANLMS_FILTER)                  += af_anlms.o
 OBJS-$(CONFIG_ANULL_FILTER)                  += af_anull.o
 OBJS-$(CONFIG_APAD_FILTER)                   += af_apad.o
 OBJS-$(CONFIG_APERMS_FILTER)                 += f_perms.o
diff --git a/libavfilter/af_anlms.c b/libavfilter/af_anlms.c
new file mode 100644
index 0000000000..571c0313e1
--- /dev/null
+++ b/libavfilter/af_anlms.c
@@ -0,0 +1,308 @@ 
+/*
+ * Copyright (c) 2019 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/avassert.h"
+#include "libavutil/channel_layout.h"
+#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"
+
+typedef struct AudioNLMSContext {
+    const AVClass *class;
+
+    int order;
+    float mu;
+    float eps;
+    float leakage;
+
+    int kernel_size;
+    AVFrame *offset;
+    AVFrame *delay;
+    AVFrame *coeffs;
+    AVFrame *tmp;
+
+    AVFrame *frame[2];
+
+    AVFloatDSPContext *fdsp;
+} AudioNLMSContext;
+
+#define OFFSET(x) offsetof(AudioNLMSContext, x)
+#define A AV_OPT_FLAG_AUDIO_PARAM|AV_OPT_FLAG_FILTERING_PARAM
+
+static const AVOption anlms_options[] = {
+    { "order",   "set the filter order",   OFFSET(order),   AV_OPT_TYPE_INT,   {.i64=256},  1, INT16_MAX, A },
+    { "mu",      "set the filter mu",      OFFSET(mu),      AV_OPT_TYPE_FLOAT, {.dbl=0.75}, 0, 1, A },
+    { "eps",     "set the filter eps",     OFFSET(eps),     AV_OPT_TYPE_FLOAT, {.dbl=1},    0, 1, A },
+    { "leakage", "set the filter leakage", OFFSET(leakage), AV_OPT_TYPE_FLOAT, {.dbl=0},    0, 1, A },
+    { NULL }
+};
+
+AVFILTER_DEFINE_CLASS(anlms);
+
+static int query_formats(AVFilterContext *ctx)
+{
+    AVFilterFormats *formats;
+    AVFilterChannelLayouts *layouts;
+    static const enum AVSampleFormat sample_fmts[] = {
+        AV_SAMPLE_FMT_FLTP,
+        AV_SAMPLE_FMT_NONE
+    };
+    int ret;
+
+    layouts = ff_all_channel_counts();
+    if (!layouts)
+        return AVERROR(ENOMEM);
+    ret = ff_set_common_channel_layouts(ctx, layouts);
+    if (ret < 0)
+        return ret;
+
+    formats = ff_make_format_list(sample_fmts);
+    if (!formats)
+        return AVERROR(ENOMEM);
+    ret = ff_set_common_formats(ctx, formats);
+    if (ret < 0)
+        return ret;
+
+    formats = ff_all_samplerates();
+    if (!formats)
+        return AVERROR(ENOMEM);
+    return ff_set_common_samplerates(ctx, formats);
+}
+
+static float fir_sample(AudioNLMSContext *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(AudioNLMSContext *s, float input, float desired,
+                            float *delay, float *coeffs, float *tmp, int *offsetp)
+{
+    const int order = s->order;
+    const float leakage = s->leakage;
+    const float mu = s->mu;
+    const float a = 1.f - leakage * mu;
+    float sum, y, e, norm, b;
+    int offset = *offsetp;
+
+    delay[offset + order] = input;
+
+    y = fir_sample(s, input, delay, coeffs, tmp, offsetp);
+    e = desired - y;
+
+    sum = s->fdsp->scalarproduct_float(delay, delay, s->kernel_size);
+
+    norm = s->eps + sum;
+    b = mu * e / norm;
+
+    memcpy(tmp, delay + offset, order * sizeof(float));
+
+    s->fdsp->vector_fmul_scalar(coeffs, coeffs, a, s->kernel_size);
+
+    s->fdsp->vector_fmac_scalar(coeffs, tmp, b, s->kernel_size);
+
+    memcpy(coeffs + order, coeffs, order * sizeof(float));
+
+    return y;
+}
+
+static int process_channels(AVFilterContext *ctx, void *arg, int jobnr, int nb_jobs)
+{
+    AudioNLMSContext *s = ctx->priv;
+    AVFrame *out = arg;
+    const int start = (out->channels * jobnr) / nb_jobs;
+    const int end = (out->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 *delay = (float *)s->delay->extended_data[c];
+        float *coeffs = (float *)s->coeffs->extended_data[c];
+        float *tmp = (float *)s->tmp->extended_data[c];
+        int *offset = (int *)s->offset->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], delay, coeffs, tmp, offset);
+    }
+
+    return 0;
+}
+
+static int activate(AVFilterContext *ctx)
+{
+    AudioNLMSContext *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);
+        }
+
+        ctx->internal->execute(ctx, process_channels, out, NULL, FFMIN(ctx->outputs[0]->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;
+    AudioNLMSContext *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->tmp)
+        s->tmp = ff_get_audio_buffer(outlink, s->kernel_size);
+    if (!s->delay || !s->coeffs || !s->offset || !s->tmp)
+        return AVERROR(ENOMEM);
+
+    return 0;
+}
+
+static av_cold int init(AVFilterContext *ctx)
+{
+    AudioNLMSContext *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)
+{
+    AudioNLMSContext *s = ctx->priv;
+
+    av_freep(&s->fdsp);
+    av_frame_free(&s->delay);
+    av_frame_free(&s->coeffs);
+    av_frame_free(&s->offset);
+    av_frame_free(&s->tmp);
+}
+
+static const AVFilterPad inputs[] = {
+    {
+        .name = "input",
+        .type = AVMEDIA_TYPE_AUDIO,
+    },
+    {
+        .name = "desired",
+        .type = AVMEDIA_TYPE_AUDIO,
+    },
+    { NULL }
+};
+
+static const AVFilterPad outputs[] = {
+    {
+        .name         = "default",
+        .type         = AVMEDIA_TYPE_AUDIO,
+        .config_props = config_output,
+    },
+    { NULL }
+};
+
+AVFilter ff_af_anlms = {
+    .name           = "anlms",
+    .description    = NULL_IF_CONFIG_SMALL("Apply Normalized Least-Mean-Squares algorithm to first audio stream."),
+    .priv_size      = sizeof(AudioNLMSContext),
+    .priv_class     = &anlms_class,
+    .init           = init,
+    .uninit         = uninit,
+    .activate       = activate,
+    .query_formats  = query_formats,
+    .inputs         = inputs,
+    .outputs        = outputs,
+    .flags          = AVFILTER_FLAG_SLICE_THREADS,
+};
diff --git a/libavfilter/allfilters.c b/libavfilter/allfilters.c
index 1a26129069..4f8b3039ed 100644
--- a/libavfilter/allfilters.c
+++ b/libavfilter/allfilters.c
@@ -56,6 +56,7 @@  extern AVFilter ff_af_amix;
 extern AVFilter ff_af_amultiply;
 extern AVFilter ff_af_anequalizer;
 extern AVFilter ff_af_anlmdn;
+extern AVFilter ff_af_anlms;
 extern AVFilter ff_af_anull;
 extern AVFilter ff_af_apad;
 extern AVFilter ff_af_aperms;