diff mbox series

[FFmpeg-devel,RFC] avfilter: add speechnorm filter

Message ID 20200503191601.15344-1-onemda@gmail.com
State New
Headers show
Series [FFmpeg-devel,RFC] avfilter: add speechnorm filter
Related show

Checks

Context Check Description
andriy/default pending
andriy/make success Make finished
andriy/make_fate success Make fate finished

Commit Message

Paul B Mahol May 3, 2020, 7:16 p.m. UTC
Signed-off-by: Paul B Mahol <onemda@gmail.com>
---
 libavfilter/Makefile        |   1 +
 libavfilter/af_speechnorm.c | 381 ++++++++++++++++++++++++++++++++++++
 libavfilter/allfilters.c    |   1 +
 3 files changed, 383 insertions(+)
 create mode 100644 libavfilter/af_speechnorm.c

Comments

Andreas Rheinhardt May 4, 2020, 6:36 a.m. UTC | #1
Paul B Mahol:
> Signed-off-by: Paul B Mahol <onemda@gmail.com>
> ---
>  libavfilter/Makefile        |   1 +
>  libavfilter/af_speechnorm.c | 381 ++++++++++++++++++++++++++++++++++++
>  libavfilter/allfilters.c    |   1 +
>  3 files changed, 383 insertions(+)
>  create mode 100644 libavfilter/af_speechnorm.c
> 
> diff --git a/libavfilter/Makefile b/libavfilter/Makefile
> index f982afe15f..421a01753e 100644
> --- a/libavfilter/Makefile
> +++ b/libavfilter/Makefile
> @@ -134,6 +134,7 @@ OBJS-$(CONFIG_SIDECHAINGATE_FILTER)          += af_agate.o
>  OBJS-$(CONFIG_SILENCEDETECT_FILTER)          += af_silencedetect.o
>  OBJS-$(CONFIG_SILENCEREMOVE_FILTER)          += af_silenceremove.o
>  OBJS-$(CONFIG_SOFALIZER_FILTER)              += af_sofalizer.o
> +OBJS-$(CONFIG_SPEECHNORM_FILTER)             += af_speechnorm.o
>  OBJS-$(CONFIG_STEREOTOOLS_FILTER)            += af_stereotools.o
>  OBJS-$(CONFIG_STEREOWIDEN_FILTER)            += af_stereowiden.o
>  OBJS-$(CONFIG_SUPEREQUALIZER_FILTER)         += af_superequalizer.o
> diff --git a/libavfilter/af_speechnorm.c b/libavfilter/af_speechnorm.c
> new file mode 100644
> index 0000000000..52fc8e6e42
> --- /dev/null
> +++ b/libavfilter/af_speechnorm.c
> @@ -0,0 +1,381 @@
> +/*
> + * Speech Normalizer
> + * Copyright (c) 2020 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
> + */
> +
> +/**
> + * @file
> + * Speech Normalizer
> + */
> +
> +#include <float.h>
> +
> +#include "libavutil/avassert.h"
> +#include "libavutil/opt.h"
> +
> +#define FF_BUFQUEUE_SIZE (1024)
> +#include "bufferqueue.h"
> +
> +#include "audio.h"
> +#include "avfilter.h"
> +#include "filters.h"
> +#include "internal.h"
> +
> +#define MAX_ITEMS  882000
> +
> +typedef struct PeriodItem {
> +    int size;
> +    int type;
> +    double max_peak;
> +} PeriodItem;
> +
> +typedef struct SpeechNormalizerContext {
> +    const AVClass *class;
> +
> +    double peak_value;
> +    double max_amplification;
> +    double threshold_value;
> +    double feedback;
> +    double decay;
> +    int channels;
> +
> +    int max_period;
> +    int eof;
> +    int64_t pts;
> +    int state[12];
> +
> +    PeriodItem pi[12][MAX_ITEMS];
> +    double gain_state[12];
> +    int pi_start[12];
> +    int pi_end[12];
> +
> +    struct FFBufQueue queue;
> +} SpeechNormalizerContext;
> +
> +#define OFFSET(x) offsetof(SpeechNormalizerContext, x)
> +#define FLAGS AV_OPT_FLAG_AUDIO_PARAM|AV_OPT_FLAG_FILTERING_PARAM|AV_OPT_FLAG_RUNTIME_PARAM
> +
> +static const AVOption speechnorm_options[] = {
> +    { "peak",        "set the peak value",          OFFSET(peak_value),          AV_OPT_TYPE_DOUBLE, {.dbl = 0.95},  0.0,     1.0, FLAGS },
> +    { "p",           "set the peak value",          OFFSET(peak_value),          AV_OPT_TYPE_DOUBLE, {.dbl = 0.95},  0.0,     1.0, FLAGS },
> +    { "maxgain",     "set the max amplification",   OFFSET(max_amplification),   AV_OPT_TYPE_DOUBLE, {.dbl = 2.0},   1.0,    10.0, FLAGS },
> +    { "m",           "set the max amplification",   OFFSET(max_amplification),   AV_OPT_TYPE_DOUBLE, {.dbl = 2.0},   1.0,    10.0, FLAGS },
> +    { "threshold",   "set the threshold value",     OFFSET(threshold_value),     AV_OPT_TYPE_DOUBLE, {.dbl = 0},     0.0,     1.0, FLAGS },
> +    { "t",           "set the threshold value",     OFFSET(threshold_value),     AV_OPT_TYPE_DOUBLE, {.dbl = 0},     0.0,     1.0, FLAGS },
> +    { "feedback",    "set the feedback value",      OFFSET(feedback),            AV_OPT_TYPE_DOUBLE, {.dbl = 0.001}, 0.0,     1.0, FLAGS },
> +    { "f",           "set the feedback value",      OFFSET(feedback),            AV_OPT_TYPE_DOUBLE, {.dbl = 0.001}, 0.0,     1.0, FLAGS },
> +    { "decay",       "set the decay value",         OFFSET(decay),               AV_OPT_TYPE_DOUBLE, {.dbl = 0.999}, 0.0,     1.0, FLAGS },
> +    { "d",           "set the decay value",         OFFSET(decay),               AV_OPT_TYPE_DOUBLE, {.dbl = 0.999}, 0.0,     1.0, FLAGS },
> +    { NULL }
> +};
> +
> +AVFILTER_DEFINE_CLASS(speechnorm);
> +
> +static int query_formats(AVFilterContext *ctx)
> +{
> +    AVFilterFormats *formats;
> +    AVFilterChannelLayouts *layouts;
> +    static const enum AVSampleFormat sample_fmts[] = {
> +        AV_SAMPLE_FMT_DBLP,
> +        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 int config_input(AVFilterLink *inlink)
> +{
> +    AVFilterContext *ctx = inlink->dst;
> +    SpeechNormalizerContext *s = ctx->priv;
> +
> +    s->max_period = inlink->sample_rate / (2 * 20);
> +    s->channels = inlink->channels;
> +    for (int ch = 0; ch < s->channels; ch++)
> +        s->state[ch] = -1;
> +
> +    return 0;
> +}
> +
> +static int get_pi_samples(PeriodItem *pi, int start, int end, int mode)
> +{
> +    int sum;
> +
> +    if (mode && pi[start].type == 0)
> +        return 0;
> +
> +    sum = pi[start].size;
> +    av_assert0(sum > 0);
> +    while (start != end) {
> +        start++;
> +        if (start >= MAX_ITEMS)
> +            start = 0;
> +        if (mode && pi[start].type == 0)
> +            break;
> +        av_assert0(pi[start].size > 0);
> +        sum += pi[start].size;
> +        if (pi[start].type == 0)
> +            break;
> +    }
> +
> +    return sum;
> +}
> +
> +static int consume_pi(PeriodItem *pi, int start, int end, int nb_samples)
> +{
> +    int sum;
> +
> +    sum = pi[start].size;
> +    av_assert0(pi[start].size > 0);
> +    while (sum < nb_samples) {
> +        av_assert0(pi[start].type == 1);
> +        av_assert0(start != end);
> +        start++;
> +        if (start >= MAX_ITEMS)
> +            start = 0;
> +        av_assert0(pi[start].size > 0);
> +        sum += pi[start].size;
> +    }
> +
> +    av_assert0(pi[start].size >= sum - nb_samples);
> +    pi[start].size = sum - nb_samples;
> +    av_assert0(pi[start].size >= 0);
> +    if (pi[start].size == 0 && start != end) {
> +        start++;
> +        if (start >= MAX_ITEMS)
> +            start = 0;
> +    }
> +
> +    return start;
> +}
> +
> +static int get_queued_samples(SpeechNormalizerContext *s)
> +{
> +    int sum = 0;
> +
> +    for (int i = 0; i < s->queue.available; i++) {
> +        AVFrame *frame = ff_bufqueue_peek(&s->queue, i);
> +        sum += frame->nb_samples;
> +    }
> +
> +    return sum;
> +}
> +
> +static int filter_frame(AVFilterContext *ctx)
> +{
> +    SpeechNormalizerContext *s = ctx->priv;
> +    AVFilterLink *outlink = ctx->outputs[0];
> +    AVFilterLink *inlink = ctx->inputs[0];
> +    int min_pi_nb_samples;
> +    AVFrame *in = NULL;
> +    int ret;
> +
> +    for (int f = 0; f < ff_inlink_queued_frames(inlink); f++) {
> +        ret = ff_inlink_consume_frame(inlink, &in);
> +        if (ret < 0)
> +            return ret;
> +        if (ret == 0)
> +            break;
> +
> +        ff_bufqueue_add(ctx, &s->queue, in);
> +
> +        for (int ch = 0; ch < inlink->channels; ch++) {
> +            const double *src = (const double *)in->extended_data[ch];
> +            int n = 0;
> +
> +            if (s->state[ch] < 0)
> +                s->state[ch] = src[0] >= 0.;
> +
> +            while (n < in->nb_samples) {
> +                if (s->state[ch] != (src[n] >= 0.) || s->pi[ch][s->pi_end[ch]].size > s->max_period) {
> +                    s->state[ch] = src[n] >= 0.;
> +                    av_assert0(s->pi[ch][s->pi_end[ch]].size > 0);
> +                    s->pi[ch][s->pi_end[ch]].type = 1;
> +                    s->pi_end[ch]++;
> +                    if (s->pi_end[ch] >= MAX_ITEMS)
> +                        s->pi_end[ch] = 0;
> +                    s->pi[ch][s->pi_end[ch]].max_peak = DBL_MIN;
> +                    s->pi[ch][s->pi_end[ch]].type = 0;
> +                    s->pi[ch][s->pi_end[ch]].size = 0;
> +                    av_assert0(s->pi_end[ch] != s->pi_start[ch]);
> +                }
> +
> +                if (src[n] >= 0.) {
> +                    while (src[n] >= 0.) {
> +                        s->pi[ch][s->pi_end[ch]].max_peak = FFMAX(s->pi[ch][s->pi_end[ch]].max_peak, FFABS(src[n]));
> +                        s->pi[ch][s->pi_end[ch]].size++;
> +                        n++;
> +                        if (n >= in->nb_samples)
> +                            break;
> +                    }
> +                } else {
> +                    while (src[n] < 0.) {
> +                        s->pi[ch][s->pi_end[ch]].max_peak = FFMAX(s->pi[ch][s->pi_end[ch]].max_peak, FFABS(src[n]));
> +                        s->pi[ch][s->pi_end[ch]].size++;
> +                        n++;
> +                        if (n >= in->nb_samples)
> +                            break;
> +                    }
> +                }
> +            }
> +        }
> +    }
> +
> +    if (s->queue.available > 0) {
> +        in = ff_bufqueue_peek(&s->queue, 0);
> +        if (!in)
> +            return 1;
> +    } else {
> +        return 1;
> +    }
> +
> +    min_pi_nb_samples = get_pi_samples(s->pi[0], s->pi_start[0], s->pi_end[0], 1);
> +    for (int ch = 1; ch < inlink->channels; ch++) {
> +        min_pi_nb_samples = FFMIN(min_pi_nb_samples, get_pi_samples(s->pi[ch], s->pi_start[ch], s->pi_end[ch], 1));
> +    }
> +
> +    if (min_pi_nb_samples >= in->nb_samples) {
> +        int nb_samples = get_queued_samples(s);
> +
> +        in = ff_bufqueue_get(&s->queue);
> +
> +        av_frame_make_writable(in);
> +
> +        nb_samples -= in->nb_samples;
> +
> +        for (int ch = 0; ch < inlink->channels; ch++) {
> +            double *src = (double *)in->extended_data[ch];
> +            int start = s->pi_start[ch];
> +            int offset = 0;
> +            double gain;
> +
> +            for (int n = 0; n < in->nb_samples; n++) {
> +                if (n >= offset) {
> +                    if (s->pi[ch][start].max_peak > s->threshold_value)
> +                        gain = FFMIN(s->max_amplification, s->peak_value / s->pi[ch][start].max_peak);
> +                    else
> +                        gain = 1.;
> +                    av_assert0(s->pi[ch][start].size > 0);
> +                    offset += s->pi[ch][start++].size;
> +                    if (start >= MAX_ITEMS)
> +                        start = 0;
> +                }
> +                s->gain_state[ch] = FFMIN(gain, gain * s->feedback + s->gain_state[ch] * s->decay);
> +                src[n] *= s->gain_state[ch];
> +            }
> +        }
> +
> +        for (int ch = 0; ch < inlink->channels; ch++) {
> +            s->pi_start[ch] = consume_pi(s->pi[ch], s->pi_start[ch], s->pi_end[ch], in->nb_samples);
> +        }
> +
> +        for (int ch = 0; ch < inlink->channels; ch++) {
> +            int pi_nb_samples = get_pi_samples(s->pi[ch], s->pi_start[ch], s->pi_end[ch], 0);
> +
> +            if (nb_samples != pi_nb_samples) {
> +                av_assert0(0);
> +            }
> +        }
> +
> +        return ff_filter_frame(outlink, in);
> +    }
> +
> +    return 1;
> +}
> +
> +static int activate(AVFilterContext *ctx)
> +{
> +    AVFilterLink *inlink = ctx->inputs[0];
> +    AVFilterLink *outlink = ctx->outputs[0];
> +    SpeechNormalizerContext *s = ctx->priv;
> +    int ret = 0, status;
> +    int64_t pts;
> +
> +    FF_FILTER_FORWARD_STATUS_BACK(outlink, inlink);
> +
> +    ret = filter_frame(ctx);
> +    if (ret <= 0)
> +        return ret;
> +
> +    if (!s->eof && ff_inlink_acknowledge_status(inlink, &status, &pts)) {
> +        if (status == AVERROR_EOF)
> +            s->eof = 1;
> +    }
> +
> +    if (s->eof && ff_inlink_queued_samples(inlink) == 0) {
> +        ff_outlink_set_status(outlink, AVERROR_EOF, s->pts);
> +        return 0;
> +    }
> +
> +    if (!s->eof)
> +        FF_FILTER_FORWARD_WANTED(outlink, inlink);
> +
> +    return FFERROR_NOT_READY;
> +}
> +
> +static av_cold void uninit(AVFilterContext *ctx)
> +{
> +}

Are AVFilters required to have an uninit function even if empty?
And is it really certain that the FFBufQueue is empty when uninit is called?

- Andreas
diff mbox series

Patch

diff --git a/libavfilter/Makefile b/libavfilter/Makefile
index f982afe15f..421a01753e 100644
--- a/libavfilter/Makefile
+++ b/libavfilter/Makefile
@@ -134,6 +134,7 @@  OBJS-$(CONFIG_SIDECHAINGATE_FILTER)          += af_agate.o
 OBJS-$(CONFIG_SILENCEDETECT_FILTER)          += af_silencedetect.o
 OBJS-$(CONFIG_SILENCEREMOVE_FILTER)          += af_silenceremove.o
 OBJS-$(CONFIG_SOFALIZER_FILTER)              += af_sofalizer.o
+OBJS-$(CONFIG_SPEECHNORM_FILTER)             += af_speechnorm.o
 OBJS-$(CONFIG_STEREOTOOLS_FILTER)            += af_stereotools.o
 OBJS-$(CONFIG_STEREOWIDEN_FILTER)            += af_stereowiden.o
 OBJS-$(CONFIG_SUPEREQUALIZER_FILTER)         += af_superequalizer.o
diff --git a/libavfilter/af_speechnorm.c b/libavfilter/af_speechnorm.c
new file mode 100644
index 0000000000..52fc8e6e42
--- /dev/null
+++ b/libavfilter/af_speechnorm.c
@@ -0,0 +1,381 @@ 
+/*
+ * Speech Normalizer
+ * Copyright (c) 2020 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
+ */
+
+/**
+ * @file
+ * Speech Normalizer
+ */
+
+#include <float.h>
+
+#include "libavutil/avassert.h"
+#include "libavutil/opt.h"
+
+#define FF_BUFQUEUE_SIZE (1024)
+#include "bufferqueue.h"
+
+#include "audio.h"
+#include "avfilter.h"
+#include "filters.h"
+#include "internal.h"
+
+#define MAX_ITEMS  882000
+
+typedef struct PeriodItem {
+    int size;
+    int type;
+    double max_peak;
+} PeriodItem;
+
+typedef struct SpeechNormalizerContext {
+    const AVClass *class;
+
+    double peak_value;
+    double max_amplification;
+    double threshold_value;
+    double feedback;
+    double decay;
+    int channels;
+
+    int max_period;
+    int eof;
+    int64_t pts;
+    int state[12];
+
+    PeriodItem pi[12][MAX_ITEMS];
+    double gain_state[12];
+    int pi_start[12];
+    int pi_end[12];
+
+    struct FFBufQueue queue;
+} SpeechNormalizerContext;
+
+#define OFFSET(x) offsetof(SpeechNormalizerContext, x)
+#define FLAGS AV_OPT_FLAG_AUDIO_PARAM|AV_OPT_FLAG_FILTERING_PARAM|AV_OPT_FLAG_RUNTIME_PARAM
+
+static const AVOption speechnorm_options[] = {
+    { "peak",        "set the peak value",          OFFSET(peak_value),          AV_OPT_TYPE_DOUBLE, {.dbl = 0.95},  0.0,     1.0, FLAGS },
+    { "p",           "set the peak value",          OFFSET(peak_value),          AV_OPT_TYPE_DOUBLE, {.dbl = 0.95},  0.0,     1.0, FLAGS },
+    { "maxgain",     "set the max amplification",   OFFSET(max_amplification),   AV_OPT_TYPE_DOUBLE, {.dbl = 2.0},   1.0,    10.0, FLAGS },
+    { "m",           "set the max amplification",   OFFSET(max_amplification),   AV_OPT_TYPE_DOUBLE, {.dbl = 2.0},   1.0,    10.0, FLAGS },
+    { "threshold",   "set the threshold value",     OFFSET(threshold_value),     AV_OPT_TYPE_DOUBLE, {.dbl = 0},     0.0,     1.0, FLAGS },
+    { "t",           "set the threshold value",     OFFSET(threshold_value),     AV_OPT_TYPE_DOUBLE, {.dbl = 0},     0.0,     1.0, FLAGS },
+    { "feedback",    "set the feedback value",      OFFSET(feedback),            AV_OPT_TYPE_DOUBLE, {.dbl = 0.001}, 0.0,     1.0, FLAGS },
+    { "f",           "set the feedback value",      OFFSET(feedback),            AV_OPT_TYPE_DOUBLE, {.dbl = 0.001}, 0.0,     1.0, FLAGS },
+    { "decay",       "set the decay value",         OFFSET(decay),               AV_OPT_TYPE_DOUBLE, {.dbl = 0.999}, 0.0,     1.0, FLAGS },
+    { "d",           "set the decay value",         OFFSET(decay),               AV_OPT_TYPE_DOUBLE, {.dbl = 0.999}, 0.0,     1.0, FLAGS },
+    { NULL }
+};
+
+AVFILTER_DEFINE_CLASS(speechnorm);
+
+static int query_formats(AVFilterContext *ctx)
+{
+    AVFilterFormats *formats;
+    AVFilterChannelLayouts *layouts;
+    static const enum AVSampleFormat sample_fmts[] = {
+        AV_SAMPLE_FMT_DBLP,
+        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 int config_input(AVFilterLink *inlink)
+{
+    AVFilterContext *ctx = inlink->dst;
+    SpeechNormalizerContext *s = ctx->priv;
+
+    s->max_period = inlink->sample_rate / (2 * 20);
+    s->channels = inlink->channels;
+    for (int ch = 0; ch < s->channels; ch++)
+        s->state[ch] = -1;
+
+    return 0;
+}
+
+static int get_pi_samples(PeriodItem *pi, int start, int end, int mode)
+{
+    int sum;
+
+    if (mode && pi[start].type == 0)
+        return 0;
+
+    sum = pi[start].size;
+    av_assert0(sum > 0);
+    while (start != end) {
+        start++;
+        if (start >= MAX_ITEMS)
+            start = 0;
+        if (mode && pi[start].type == 0)
+            break;
+        av_assert0(pi[start].size > 0);
+        sum += pi[start].size;
+        if (pi[start].type == 0)
+            break;
+    }
+
+    return sum;
+}
+
+static int consume_pi(PeriodItem *pi, int start, int end, int nb_samples)
+{
+    int sum;
+
+    sum = pi[start].size;
+    av_assert0(pi[start].size > 0);
+    while (sum < nb_samples) {
+        av_assert0(pi[start].type == 1);
+        av_assert0(start != end);
+        start++;
+        if (start >= MAX_ITEMS)
+            start = 0;
+        av_assert0(pi[start].size > 0);
+        sum += pi[start].size;
+    }
+
+    av_assert0(pi[start].size >= sum - nb_samples);
+    pi[start].size = sum - nb_samples;
+    av_assert0(pi[start].size >= 0);
+    if (pi[start].size == 0 && start != end) {
+        start++;
+        if (start >= MAX_ITEMS)
+            start = 0;
+    }
+
+    return start;
+}
+
+static int get_queued_samples(SpeechNormalizerContext *s)
+{
+    int sum = 0;
+
+    for (int i = 0; i < s->queue.available; i++) {
+        AVFrame *frame = ff_bufqueue_peek(&s->queue, i);
+        sum += frame->nb_samples;
+    }
+
+    return sum;
+}
+
+static int filter_frame(AVFilterContext *ctx)
+{
+    SpeechNormalizerContext *s = ctx->priv;
+    AVFilterLink *outlink = ctx->outputs[0];
+    AVFilterLink *inlink = ctx->inputs[0];
+    int min_pi_nb_samples;
+    AVFrame *in = NULL;
+    int ret;
+
+    for (int f = 0; f < ff_inlink_queued_frames(inlink); f++) {
+        ret = ff_inlink_consume_frame(inlink, &in);
+        if (ret < 0)
+            return ret;
+        if (ret == 0)
+            break;
+
+        ff_bufqueue_add(ctx, &s->queue, in);
+
+        for (int ch = 0; ch < inlink->channels; ch++) {
+            const double *src = (const double *)in->extended_data[ch];
+            int n = 0;
+
+            if (s->state[ch] < 0)
+                s->state[ch] = src[0] >= 0.;
+
+            while (n < in->nb_samples) {
+                if (s->state[ch] != (src[n] >= 0.) || s->pi[ch][s->pi_end[ch]].size > s->max_period) {
+                    s->state[ch] = src[n] >= 0.;
+                    av_assert0(s->pi[ch][s->pi_end[ch]].size > 0);
+                    s->pi[ch][s->pi_end[ch]].type = 1;
+                    s->pi_end[ch]++;
+                    if (s->pi_end[ch] >= MAX_ITEMS)
+                        s->pi_end[ch] = 0;
+                    s->pi[ch][s->pi_end[ch]].max_peak = DBL_MIN;
+                    s->pi[ch][s->pi_end[ch]].type = 0;
+                    s->pi[ch][s->pi_end[ch]].size = 0;
+                    av_assert0(s->pi_end[ch] != s->pi_start[ch]);
+                }
+
+                if (src[n] >= 0.) {
+                    while (src[n] >= 0.) {
+                        s->pi[ch][s->pi_end[ch]].max_peak = FFMAX(s->pi[ch][s->pi_end[ch]].max_peak, FFABS(src[n]));
+                        s->pi[ch][s->pi_end[ch]].size++;
+                        n++;
+                        if (n >= in->nb_samples)
+                            break;
+                    }
+                } else {
+                    while (src[n] < 0.) {
+                        s->pi[ch][s->pi_end[ch]].max_peak = FFMAX(s->pi[ch][s->pi_end[ch]].max_peak, FFABS(src[n]));
+                        s->pi[ch][s->pi_end[ch]].size++;
+                        n++;
+                        if (n >= in->nb_samples)
+                            break;
+                    }
+                }
+            }
+        }
+    }
+
+    if (s->queue.available > 0) {
+        in = ff_bufqueue_peek(&s->queue, 0);
+        if (!in)
+            return 1;
+    } else {
+        return 1;
+    }
+
+    min_pi_nb_samples = get_pi_samples(s->pi[0], s->pi_start[0], s->pi_end[0], 1);
+    for (int ch = 1; ch < inlink->channels; ch++) {
+        min_pi_nb_samples = FFMIN(min_pi_nb_samples, get_pi_samples(s->pi[ch], s->pi_start[ch], s->pi_end[ch], 1));
+    }
+
+    if (min_pi_nb_samples >= in->nb_samples) {
+        int nb_samples = get_queued_samples(s);
+
+        in = ff_bufqueue_get(&s->queue);
+
+        av_frame_make_writable(in);
+
+        nb_samples -= in->nb_samples;
+
+        for (int ch = 0; ch < inlink->channels; ch++) {
+            double *src = (double *)in->extended_data[ch];
+            int start = s->pi_start[ch];
+            int offset = 0;
+            double gain;
+
+            for (int n = 0; n < in->nb_samples; n++) {
+                if (n >= offset) {
+                    if (s->pi[ch][start].max_peak > s->threshold_value)
+                        gain = FFMIN(s->max_amplification, s->peak_value / s->pi[ch][start].max_peak);
+                    else
+                        gain = 1.;
+                    av_assert0(s->pi[ch][start].size > 0);
+                    offset += s->pi[ch][start++].size;
+                    if (start >= MAX_ITEMS)
+                        start = 0;
+                }
+                s->gain_state[ch] = FFMIN(gain, gain * s->feedback + s->gain_state[ch] * s->decay);
+                src[n] *= s->gain_state[ch];
+            }
+        }
+
+        for (int ch = 0; ch < inlink->channels; ch++) {
+            s->pi_start[ch] = consume_pi(s->pi[ch], s->pi_start[ch], s->pi_end[ch], in->nb_samples);
+        }
+
+        for (int ch = 0; ch < inlink->channels; ch++) {
+            int pi_nb_samples = get_pi_samples(s->pi[ch], s->pi_start[ch], s->pi_end[ch], 0);
+
+            if (nb_samples != pi_nb_samples) {
+                av_assert0(0);
+            }
+        }
+
+        return ff_filter_frame(outlink, in);
+    }
+
+    return 1;
+}
+
+static int activate(AVFilterContext *ctx)
+{
+    AVFilterLink *inlink = ctx->inputs[0];
+    AVFilterLink *outlink = ctx->outputs[0];
+    SpeechNormalizerContext *s = ctx->priv;
+    int ret = 0, status;
+    int64_t pts;
+
+    FF_FILTER_FORWARD_STATUS_BACK(outlink, inlink);
+
+    ret = filter_frame(ctx);
+    if (ret <= 0)
+        return ret;
+
+    if (!s->eof && ff_inlink_acknowledge_status(inlink, &status, &pts)) {
+        if (status == AVERROR_EOF)
+            s->eof = 1;
+    }
+
+    if (s->eof && ff_inlink_queued_samples(inlink) == 0) {
+        ff_outlink_set_status(outlink, AVERROR_EOF, s->pts);
+        return 0;
+    }
+
+    if (!s->eof)
+        FF_FILTER_FORWARD_WANTED(outlink, inlink);
+
+    return FFERROR_NOT_READY;
+}
+
+static av_cold void uninit(AVFilterContext *ctx)
+{
+}
+
+static const AVFilterPad avfilter_af_speechnorm_inputs[] = {
+    {
+        .name           = "default",
+        .type           = AVMEDIA_TYPE_AUDIO,
+        .config_props   = config_input,
+    },
+    { NULL }
+};
+
+static const AVFilterPad avfilter_af_speechnorm_outputs[] = {
+    {
+        .name          = "default",
+        .type          = AVMEDIA_TYPE_AUDIO,
+    },
+    { NULL }
+};
+
+AVFilter ff_af_speechnorm = {
+    .name          = "speechnorm",
+    .description   = NULL_IF_CONFIG_SMALL("Speech Normalizer."),
+    .query_formats = query_formats,
+    .priv_size     = sizeof(SpeechNormalizerContext),
+    .priv_class    = &speechnorm_class,
+    .activate      = activate,
+    .uninit        = uninit,
+    .inputs        = avfilter_af_speechnorm_inputs,
+    .outputs       = avfilter_af_speechnorm_outputs,
+    .process_command = ff_filter_process_command,
+};
diff --git a/libavfilter/allfilters.c b/libavfilter/allfilters.c
index 1b94501da0..fbe9633a99 100644
--- a/libavfilter/allfilters.c
+++ b/libavfilter/allfilters.c
@@ -128,6 +128,7 @@  extern AVFilter ff_af_sidechaingate;
 extern AVFilter ff_af_silencedetect;
 extern AVFilter ff_af_silenceremove;
 extern AVFilter ff_af_sofalizer;
+extern AVFilter ff_af_speechnorm;
 extern AVFilter ff_af_stereotools;
 extern AVFilter ff_af_stereowiden;
 extern AVFilter ff_af_superequalizer;