lavfi: add volume filter
authorJustin Ruggles <justin.ruggles@gmail.com>
Sat, 29 Sep 2012 04:38:13 +0000 (00:38 -0400)
committerJustin Ruggles <justin.ruggles@gmail.com>
Wed, 5 Dec 2012 16:23:37 +0000 (11:23 -0500)
Based on the volume filter in FFmpeg written by Stefano Sabatini
<stefasab@gmail.com>.

Changelog
doc/filters.texi
libavfilter/Makefile
libavfilter/af_volume.c [new file with mode: 0644]
libavfilter/af_volume.h [new file with mode: 0644]
libavfilter/allfilters.c
libavfilter/version.h

index 33e55a5..7216622 100644 (file)
--- a/Changelog
+++ b/Changelog
@@ -4,6 +4,7 @@ releases are sorted from youngest to oldest.
 version <next>:
 - ashowinfo audio filter
 - 24-bit FLAC encoding
+- audio volume filter
 
 
 version 9_beta2:
index f092f3c..55e4468 100644 (file)
@@ -359,6 +359,59 @@ not meant to be used directly, it is inserted automatically by libavfilter
 whenever conversion is needed. Use the @var{aformat} filter to force a specific
 conversion.
 
+@section volume
+
+Adjust the input audio volume.
+
+The filter accepts the following named parameters:
+@table @option
+
+@item volume
+Expresses how the audio volume will be increased or decreased.
+
+Output values are clipped to the maximum value.
+
+The output audio volume is given by the relation:
+@example
+@var{output_volume} = @var{volume} * @var{input_volume}
+@end example
+
+Default value for @var{volume} is 1.0.
+
+@item precision
+Mathematical precision.
+
+This determines which input sample formats will be allowed, which affects the
+precision of the volume scaling.
+
+@table @option
+@item fixed
+8-bit fixed-point; limits input sample format to U8, S16, and S32.
+@item float
+32-bit floating-point; limits input sample format to FLT. (default)
+@item double
+64-bit floating-point; limits input sample format to DBL.
+@end table
+@end table
+
+@subsection Examples
+
+@itemize
+@item
+Halve the input audio volume:
+@example
+volume=volume=0.5
+volume=volume=1/2
+volume=volume=-6.0206dB
+@end example
+
+@item
+Increase input audio power by 6 decibels using fixed-point precision:
+@example
+volume=volume=6dB:precision=fixed
+@end example
+@end itemize
+
 @c man end AUDIO FILTERS
 
 @chapter Audio Sources
index 752ff40..2559e8a 100644 (file)
@@ -35,6 +35,7 @@ OBJS-$(CONFIG_CHANNELMAP_FILTER)             += af_channelmap.o
 OBJS-$(CONFIG_CHANNELSPLIT_FILTER)           += af_channelsplit.o
 OBJS-$(CONFIG_JOIN_FILTER)                   += af_join.o
 OBJS-$(CONFIG_RESAMPLE_FILTER)               += af_resample.o
+OBJS-$(CONFIG_VOLUME_FILTER)                 += af_volume.o
 
 OBJS-$(CONFIG_ANULLSRC_FILTER)               += asrc_anullsrc.o
 
diff --git a/libavfilter/af_volume.c b/libavfilter/af_volume.c
new file mode 100644 (file)
index 0000000..4a4e29f
--- /dev/null
@@ -0,0 +1,314 @@
+/*
+ * Copyright (c) 2011 Stefano Sabatini
+ * Copyright (c) 2012 Justin Ruggles <justin.ruggles@gmail.com>
+ *
+ * This file is part of Libav.
+ *
+ * Libav 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.
+ *
+ * Libav 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 Libav; if not, write to the Free Software
+ * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
+ */
+
+/**
+ * @file
+ * audio volume filter
+ */
+
+#include "libavutil/audioconvert.h"
+#include "libavutil/common.h"
+#include "libavutil/eval.h"
+#include "libavutil/float_dsp.h"
+#include "libavutil/opt.h"
+#include "audio.h"
+#include "avfilter.h"
+#include "formats.h"
+#include "internal.h"
+#include "af_volume.h"
+
+static const char *precision_str[] = {
+    "fixed", "float", "double"
+};
+
+#define OFFSET(x) offsetof(VolumeContext, x)
+#define A AV_OPT_FLAG_AUDIO_PARAM
+
+static const AVOption options[] = {
+    { "volume", "Volume adjustment.",
+            OFFSET(volume), AV_OPT_TYPE_DOUBLE, { .dbl = 1.0 }, 0, 0x7fffff, A },
+    { "precision", "Mathematical precision.",
+            OFFSET(precision), AV_OPT_TYPE_INT, { .i64 = PRECISION_FLOAT }, PRECISION_FIXED, PRECISION_DOUBLE, A, "precision" },
+        { "fixed",  "8-bit fixed-point.",     0, AV_OPT_TYPE_CONST, { .i64 = PRECISION_FIXED  }, INT_MIN, INT_MAX, A, "precision" },
+        { "float",  "32-bit floating-point.", 0, AV_OPT_TYPE_CONST, { .i64 = PRECISION_FLOAT  }, INT_MIN, INT_MAX, A, "precision" },
+        { "double", "64-bit floating-point.", 0, AV_OPT_TYPE_CONST, { .i64 = PRECISION_DOUBLE }, INT_MIN, INT_MAX, A, "precision" },
+    { NULL },
+};
+
+static const AVClass volume_class = {
+    .class_name = "volume filter",
+    .item_name  = av_default_item_name,
+    .option     = options,
+    .version    = LIBAVUTIL_VERSION_INT,
+};
+
+static av_cold int init(AVFilterContext *ctx, const char *args)
+{
+    VolumeContext *vol = ctx->priv;
+    int ret;
+
+    vol->class = &volume_class;
+    av_opt_set_defaults(vol);
+
+    if ((ret = av_set_options_string(vol, args, "=", ":")) < 0) {
+        av_log(ctx, AV_LOG_ERROR, "Error parsing options string '%s'.\n", args);
+        return ret;
+    }
+
+    if (vol->precision == PRECISION_FIXED) {
+        vol->volume_i = (int)(vol->volume * 256 + 0.5);
+        vol->volume   = vol->volume_i / 256.0;
+        av_log(ctx, AV_LOG_VERBOSE, "volume:(%d/256)(%f)(%1.2fdB) precision:fixed\n",
+               vol->volume_i, vol->volume, 20.0*log(vol->volume)/M_LN10);
+    } else {
+        av_log(ctx, AV_LOG_VERBOSE, "volume:(%f)(%1.2fdB) precision:%s\n",
+               vol->volume, 20.0*log(vol->volume)/M_LN10,
+               precision_str[vol->precision]);
+    }
+
+    av_opt_free(vol);
+    return ret;
+}
+
+static int query_formats(AVFilterContext *ctx)
+{
+    VolumeContext *vol = ctx->priv;
+    AVFilterFormats *formats = NULL;
+    AVFilterChannelLayouts *layouts;
+    static const enum AVSampleFormat sample_fmts[][7] = {
+        /* PRECISION_FIXED */
+        {
+            AV_SAMPLE_FMT_U8,
+            AV_SAMPLE_FMT_U8P,
+            AV_SAMPLE_FMT_S16,
+            AV_SAMPLE_FMT_S16P,
+            AV_SAMPLE_FMT_S32,
+            AV_SAMPLE_FMT_S32P,
+            AV_SAMPLE_FMT_NONE
+        },
+        /* PRECISION_FLOAT */
+        {
+            AV_SAMPLE_FMT_FLT,
+            AV_SAMPLE_FMT_FLTP,
+            AV_SAMPLE_FMT_NONE
+        },
+        /* PRECISION_DOUBLE */
+        {
+            AV_SAMPLE_FMT_DBL,
+            AV_SAMPLE_FMT_DBLP,
+            AV_SAMPLE_FMT_NONE
+        }
+    };
+
+    layouts = ff_all_channel_layouts();
+    if (!layouts)
+        return AVERROR(ENOMEM);
+    ff_set_common_channel_layouts(ctx, layouts);
+
+    formats = ff_make_format_list(sample_fmts[vol->precision]);
+    if (!formats)
+        return AVERROR(ENOMEM);
+    ff_set_common_formats(ctx, formats);
+
+    formats = ff_all_samplerates();
+    if (!formats)
+        return AVERROR(ENOMEM);
+    ff_set_common_samplerates(ctx, formats);
+
+    return 0;
+}
+
+static inline void scale_samples_u8(uint8_t *dst, const uint8_t *src,
+                                    int nb_samples, int volume)
+{
+    int i;
+    for (i = 0; i < nb_samples; i++)
+        dst[i] = av_clip_uint8(((((int64_t)src[i] - 128) * volume + 128) >> 8) + 128);
+}
+
+static inline void scale_samples_u8_small(uint8_t *dst, const uint8_t *src,
+                                          int nb_samples, int volume)
+{
+    int i;
+    for (i = 0; i < nb_samples; i++)
+        dst[i] = av_clip_uint8((((src[i] - 128) * volume + 128) >> 8) + 128);
+}
+
+static inline void scale_samples_s16(uint8_t *dst, const uint8_t *src,
+                                     int nb_samples, int volume)
+{
+    int i;
+    int16_t *smp_dst       = (int16_t *)dst;
+    const int16_t *smp_src = (const int16_t *)src;
+    for (i = 0; i < nb_samples; i++)
+        smp_dst[i] = av_clip_int16(((int64_t)smp_src[i] * volume + 128) >> 8);
+}
+
+static inline void scale_samples_s16_small(uint8_t *dst, const uint8_t *src,
+                                           int nb_samples, int volume)
+{
+    int i;
+    int16_t *smp_dst       = (int16_t *)dst;
+    const int16_t *smp_src = (const int16_t *)src;
+    for (i = 0; i < nb_samples; i++)
+        smp_dst[i] = av_clip_int16((smp_src[i] * volume + 128) >> 8);
+}
+
+static inline void scale_samples_s32(uint8_t *dst, const uint8_t *src,
+                                     int nb_samples, int volume)
+{
+    int i;
+    int32_t *smp_dst       = (int32_t *)dst;
+    const int32_t *smp_src = (const int32_t *)src;
+    for (i = 0; i < nb_samples; i++)
+        smp_dst[i] = av_clipl_int32((((int64_t)smp_src[i] * volume + 128) >> 8));
+}
+
+
+
+static void volume_init(VolumeContext *vol)
+{
+    vol->samples_align = 1;
+
+    switch (av_get_packed_sample_fmt(vol->sample_fmt)) {
+    case AV_SAMPLE_FMT_U8:
+        if (vol->volume_i < 0x1000000)
+            vol->scale_samples = scale_samples_u8_small;
+        else
+            vol->scale_samples = scale_samples_u8;
+        break;
+    case AV_SAMPLE_FMT_S16:
+        if (vol->volume_i < 0x10000)
+            vol->scale_samples = scale_samples_s16_small;
+        else
+            vol->scale_samples = scale_samples_s16;
+        break;
+    case AV_SAMPLE_FMT_S32:
+        vol->scale_samples = scale_samples_s32;
+        break;
+    case AV_SAMPLE_FMT_FLT:
+        avpriv_float_dsp_init(&vol->fdsp, 0);
+        vol->samples_align = 4;
+        break;
+    case AV_SAMPLE_FMT_DBL:
+        avpriv_float_dsp_init(&vol->fdsp, 0);
+        vol->samples_align = 8;
+        break;
+    }
+}
+
+static int config_output(AVFilterLink *outlink)
+{
+    AVFilterContext *ctx = outlink->src;
+    VolumeContext *vol   = ctx->priv;
+    AVFilterLink *inlink = ctx->inputs[0];
+
+    vol->sample_fmt = inlink->format;
+    vol->channels   = av_get_channel_layout_nb_channels(inlink->channel_layout);
+    vol->planes     = av_sample_fmt_is_planar(inlink->format) ? vol->channels : 1;
+
+    volume_init(vol);
+
+    return 0;
+}
+
+static int filter_frame(AVFilterLink *inlink, AVFilterBufferRef *buf)
+{
+    VolumeContext *vol    = inlink->dst->priv;
+    AVFilterLink *outlink = inlink->dst->outputs[0];
+    int nb_samples        = buf->audio->nb_samples;
+    AVFilterBufferRef *out_buf;
+
+    if (vol->volume == 1.0 || vol->volume_i == 256)
+        return ff_filter_frame(outlink, buf);
+
+    /* do volume scaling in-place if input buffer is writable */
+    if (buf->perms & AV_PERM_WRITE) {
+        out_buf = buf;
+    } else {
+        out_buf = ff_get_audio_buffer(inlink, AV_PERM_WRITE, nb_samples);
+        if (!out_buf)
+            return AVERROR(ENOMEM);
+        out_buf->pts = buf->pts;
+    }
+
+    if (vol->precision != PRECISION_FIXED || vol->volume_i > 0) {
+        int p, plane_samples;
+
+        if (av_sample_fmt_is_planar(buf->format))
+            plane_samples = FFALIGN(nb_samples, vol->samples_align);
+        else
+            plane_samples = FFALIGN(nb_samples * vol->channels, vol->samples_align);
+
+        if (vol->precision == PRECISION_FIXED) {
+            for (p = 0; p < vol->planes; p++) {
+                vol->scale_samples(out_buf->extended_data[p],
+                                   buf->extended_data[p], plane_samples,
+                                   vol->volume_i);
+            }
+        } else if (av_get_packed_sample_fmt(vol->sample_fmt) == AV_SAMPLE_FMT_FLT) {
+            for (p = 0; p < vol->planes; p++) {
+                vol->fdsp.vector_fmul_scalar((float *)out_buf->extended_data[p],
+                                             (const float *)buf->extended_data[p],
+                                             vol->volume, plane_samples);
+            }
+        } else {
+            for (p = 0; p < vol->planes; p++) {
+                vol->fdsp.vector_dmul_scalar((double *)out_buf->extended_data[p],
+                                             (const double *)buf->extended_data[p],
+                                             vol->volume, plane_samples);
+            }
+        }
+    }
+
+    if (buf != out_buf)
+        avfilter_unref_buffer(buf);
+
+    return ff_filter_frame(outlink, out_buf);
+}
+
+static const AVFilterPad avfilter_af_volume_inputs[] = {
+    {
+        .name           = "default",
+        .type           = AVMEDIA_TYPE_AUDIO,
+        .filter_frame   = filter_frame,
+    },
+    { NULL }
+};
+
+static const AVFilterPad avfilter_af_volume_outputs[] = {
+    {
+        .name         = "default",
+        .type         = AVMEDIA_TYPE_AUDIO,
+        .config_props = config_output,
+    },
+    { NULL }
+};
+
+AVFilter avfilter_af_volume = {
+    .name           = "volume",
+    .description    = NULL_IF_CONFIG_SMALL("Change input volume."),
+    .query_formats  = query_formats,
+    .priv_size      = sizeof(VolumeContext),
+    .init           = init,
+    .inputs         = avfilter_af_volume_inputs,
+    .outputs        = avfilter_af_volume_outputs,
+};
diff --git a/libavfilter/af_volume.h b/libavfilter/af_volume.h
new file mode 100644 (file)
index 0000000..dec8767
--- /dev/null
@@ -0,0 +1,53 @@
+/*
+ * This file is part of Libav.
+ *
+ * Libav 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.
+ *
+ * Libav 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 Libav; if not, write to the Free Software
+ * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
+ */
+
+/**
+ * @file
+ * audio volume filter
+ */
+
+#ifndef AVFILTER_AF_VOLUME_H
+#define AVFILTER_AF_VOLUME_H
+
+#include "libavutil/common.h"
+#include "libavutil/float_dsp.h"
+#include "libavutil/opt.h"
+#include "libavutil/samplefmt.h"
+
+enum PrecisionType {
+    PRECISION_FIXED = 0,
+    PRECISION_FLOAT,
+    PRECISION_DOUBLE,
+};
+
+typedef struct VolumeContext {
+    const AVClass *class;
+    AVFloatDSPContext fdsp;
+    enum PrecisionType precision;
+    double volume;
+    int    volume_i;
+    int    channels;
+    int    planes;
+    enum AVSampleFormat sample_fmt;
+
+    void (*scale_samples)(uint8_t *dst, const uint8_t *src, int nb_samples,
+                          int volume);
+    int samples_align;
+} VolumeContext;
+
+#endif /* AVFILTER_AF_VOLUME_H */
index d7a7b07..0d7cbc2 100644 (file)
@@ -46,6 +46,7 @@ void avfilter_register_all(void)
     REGISTER_FILTER (CHANNELSPLIT,channelsplit,af);
     REGISTER_FILTER (JOIN,        join,        af);
     REGISTER_FILTER (RESAMPLE,    resample,    af);
+    REGISTER_FILTER (VOLUME,      volume,      af);
 
     REGISTER_FILTER (ANULLSRC,    anullsrc,    asrc);
 
index eb5326b..c09d44b 100644 (file)
@@ -29,7 +29,7 @@
 #include "libavutil/avutil.h"
 
 #define LIBAVFILTER_VERSION_MAJOR  3
-#define LIBAVFILTER_VERSION_MINOR  2
+#define LIBAVFILTER_VERSION_MINOR  3
 #define LIBAVFILTER_VERSION_MICRO  0
 
 #define LIBAVFILTER_VERSION_INT AV_VERSION_INT(LIBAVFILTER_VERSION_MAJOR, \