mirror of
https://gitlab.freedesktop.org/gstreamer/gstreamer.git
synced 2024-12-21 15:56:42 +00:00
d8fa279161
Split plugin into features including dynamic types which can be indiviually registered during a static build. More details here: https://gitlab.freedesktop.org/gstreamer/gst-build/-/merge_requests/199 https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/661 Part-of: <https://gitlab.freedesktop.org/gstreamer/gst-plugins-good/-/merge_requests/876>
718 lines
22 KiB
C
718 lines
22 KiB
C
/*
|
|
* GStreamer
|
|
* Copyright (C) 2007 Sebastian Dröge <slomo@circular-chaos.org>
|
|
*
|
|
* This library is free software; you can redistribute it and/or
|
|
* modify it under the terms of the GNU Library General Public
|
|
* License as published by the Free Software Foundation; either
|
|
* version 2 of the License, or (at your option) any later version.
|
|
*
|
|
* This library 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
|
|
* Library General Public License for more details.
|
|
*
|
|
* You should have received a copy of the GNU Library General Public
|
|
* License along with this library; if not, write to the
|
|
* Free Software Foundation, Inc., 51 Franklin St, Fifth Floor,
|
|
* Boston, MA 02110-1301, USA.
|
|
*/
|
|
|
|
/**
|
|
* SECTION:element-audiodynamic
|
|
* @title: audiodynamic
|
|
*
|
|
* This element can act as a compressor or expander. A compressor changes the
|
|
* amplitude of all samples above a specific threshold with a specific ratio,
|
|
* a expander does the same for all samples below a specific threshold. If
|
|
* soft-knee mode is selected the ratio is applied smoothly.
|
|
*
|
|
* ## Example launch line
|
|
* |[
|
|
* gst-launch-1.0 audiotestsrc wave=saw ! audiodynamic characteristics=soft-knee mode=compressor threshold=0.5 ratio=0.5 ! alsasink
|
|
* gst-launch-1.0 filesrc location="melo1.ogg" ! oggdemux ! vorbisdec ! audioconvert ! audiodynamic characteristics=hard-knee mode=expander threshold=0.2 ratio=4.0 ! alsasink
|
|
* gst-launch-1.0 audiotestsrc wave=saw ! audioconvert ! audiodynamic ! audioconvert ! alsasink
|
|
* ]|
|
|
*
|
|
*/
|
|
|
|
/* TODO: Implement attack and release parameters */
|
|
|
|
#ifdef HAVE_CONFIG_H
|
|
#include "config.h"
|
|
#endif
|
|
|
|
#include <gst/gst.h>
|
|
#include <gst/base/gstbasetransform.h>
|
|
#include <gst/audio/audio.h>
|
|
#include <gst/audio/gstaudiofilter.h>
|
|
|
|
#include "audiodynamic.h"
|
|
|
|
#define GST_CAT_DEFAULT gst_audio_dynamic_debug
|
|
GST_DEBUG_CATEGORY_STATIC (GST_CAT_DEFAULT);
|
|
|
|
/* Filter signals and args */
|
|
enum
|
|
{
|
|
/* FILL ME */
|
|
LAST_SIGNAL
|
|
};
|
|
|
|
enum
|
|
{
|
|
PROP_0,
|
|
PROP_CHARACTERISTICS,
|
|
PROP_MODE,
|
|
PROP_THRESHOLD,
|
|
PROP_RATIO
|
|
};
|
|
|
|
#define ALLOWED_CAPS \
|
|
"audio/x-raw," \
|
|
" format=(string) {"GST_AUDIO_NE(S16)","GST_AUDIO_NE(F32)"}," \
|
|
" rate=(int)[1,MAX]," \
|
|
" channels=(int)[1,MAX]," \
|
|
" layout=(string) {interleaved, non-interleaved}"
|
|
|
|
G_DEFINE_TYPE (GstAudioDynamic, gst_audio_dynamic, GST_TYPE_AUDIO_FILTER);
|
|
GST_ELEMENT_REGISTER_DEFINE (audiodynamic, "audiodynamic",
|
|
GST_RANK_NONE, GST_TYPE_AUDIO_DYNAMIC);
|
|
|
|
static void gst_audio_dynamic_set_property (GObject * object, guint prop_id,
|
|
const GValue * value, GParamSpec * pspec);
|
|
static void gst_audio_dynamic_get_property (GObject * object, guint prop_id,
|
|
GValue * value, GParamSpec * pspec);
|
|
|
|
static gboolean gst_audio_dynamic_setup (GstAudioFilter * filter,
|
|
const GstAudioInfo * info);
|
|
static GstFlowReturn gst_audio_dynamic_transform_ip (GstBaseTransform * base,
|
|
GstBuffer * buf);
|
|
|
|
static void
|
|
gst_audio_dynamic_transform_hard_knee_compressor_int (GstAudioDynamic * filter,
|
|
gint16 * data, guint num_samples);
|
|
static void
|
|
gst_audio_dynamic_transform_hard_knee_compressor_float (GstAudioDynamic *
|
|
filter, gfloat * data, guint num_samples);
|
|
static void
|
|
gst_audio_dynamic_transform_soft_knee_compressor_int (GstAudioDynamic * filter,
|
|
gint16 * data, guint num_samples);
|
|
static void
|
|
gst_audio_dynamic_transform_soft_knee_compressor_float (GstAudioDynamic *
|
|
filter, gfloat * data, guint num_samples);
|
|
static void gst_audio_dynamic_transform_hard_knee_expander_int (GstAudioDynamic
|
|
* filter, gint16 * data, guint num_samples);
|
|
static void
|
|
gst_audio_dynamic_transform_hard_knee_expander_float (GstAudioDynamic * filter,
|
|
gfloat * data, guint num_samples);
|
|
static void gst_audio_dynamic_transform_soft_knee_expander_int (GstAudioDynamic
|
|
* filter, gint16 * data, guint num_samples);
|
|
static void
|
|
gst_audio_dynamic_transform_soft_knee_expander_float (GstAudioDynamic * filter,
|
|
gfloat * data, guint num_samples);
|
|
|
|
static const GstAudioDynamicProcessFunc process_functions[] = {
|
|
(GstAudioDynamicProcessFunc)
|
|
gst_audio_dynamic_transform_hard_knee_compressor_int,
|
|
(GstAudioDynamicProcessFunc)
|
|
gst_audio_dynamic_transform_hard_knee_compressor_float,
|
|
(GstAudioDynamicProcessFunc)
|
|
gst_audio_dynamic_transform_soft_knee_compressor_int,
|
|
(GstAudioDynamicProcessFunc)
|
|
gst_audio_dynamic_transform_soft_knee_compressor_float,
|
|
(GstAudioDynamicProcessFunc)
|
|
gst_audio_dynamic_transform_hard_knee_expander_int,
|
|
(GstAudioDynamicProcessFunc)
|
|
gst_audio_dynamic_transform_hard_knee_expander_float,
|
|
(GstAudioDynamicProcessFunc)
|
|
gst_audio_dynamic_transform_soft_knee_expander_int,
|
|
(GstAudioDynamicProcessFunc)
|
|
gst_audio_dynamic_transform_soft_knee_expander_float
|
|
};
|
|
|
|
enum
|
|
{
|
|
CHARACTERISTICS_HARD_KNEE = 0,
|
|
CHARACTERISTICS_SOFT_KNEE
|
|
};
|
|
|
|
#define GST_TYPE_AUDIO_DYNAMIC_CHARACTERISTICS (gst_audio_dynamic_characteristics_get_type ())
|
|
static GType
|
|
gst_audio_dynamic_characteristics_get_type (void)
|
|
{
|
|
static GType gtype = 0;
|
|
|
|
if (gtype == 0) {
|
|
static const GEnumValue values[] = {
|
|
{CHARACTERISTICS_HARD_KNEE, "Hard Knee (default)",
|
|
"hard-knee"},
|
|
{CHARACTERISTICS_SOFT_KNEE, "Soft Knee (smooth)",
|
|
"soft-knee"},
|
|
{0, NULL, NULL}
|
|
};
|
|
|
|
gtype = g_enum_register_static ("GstAudioDynamicCharacteristics", values);
|
|
}
|
|
return gtype;
|
|
}
|
|
|
|
enum
|
|
{
|
|
MODE_COMPRESSOR = 0,
|
|
MODE_EXPANDER
|
|
};
|
|
|
|
#define GST_TYPE_AUDIO_DYNAMIC_MODE (gst_audio_dynamic_mode_get_type ())
|
|
static GType
|
|
gst_audio_dynamic_mode_get_type (void)
|
|
{
|
|
static GType gtype = 0;
|
|
|
|
if (gtype == 0) {
|
|
static const GEnumValue values[] = {
|
|
{MODE_COMPRESSOR, "Compressor (default)",
|
|
"compressor"},
|
|
{MODE_EXPANDER, "Expander", "expander"},
|
|
{0, NULL, NULL}
|
|
};
|
|
|
|
gtype = g_enum_register_static ("GstAudioDynamicMode", values);
|
|
}
|
|
return gtype;
|
|
}
|
|
|
|
static void
|
|
gst_audio_dynamic_set_process_function (GstAudioDynamic * filter,
|
|
const GstAudioInfo * info)
|
|
{
|
|
gint func_index;
|
|
|
|
func_index = (filter->mode == MODE_COMPRESSOR) ? 0 : 4;
|
|
func_index += (filter->characteristics == CHARACTERISTICS_HARD_KNEE) ? 0 : 2;
|
|
func_index += (GST_AUDIO_INFO_FORMAT (info) == GST_AUDIO_FORMAT_F32) ? 1 : 0;
|
|
|
|
g_assert (func_index >= 0 && func_index < G_N_ELEMENTS (process_functions));
|
|
|
|
filter->process = process_functions[func_index];
|
|
}
|
|
|
|
/* GObject vmethod implementations */
|
|
|
|
static void
|
|
gst_audio_dynamic_class_init (GstAudioDynamicClass * klass)
|
|
{
|
|
GObjectClass *gobject_class;
|
|
GstElementClass *gstelement_class;
|
|
GstCaps *caps;
|
|
|
|
GST_DEBUG_CATEGORY_INIT (gst_audio_dynamic_debug, "audiodynamic", 0,
|
|
"audiodynamic element");
|
|
|
|
gobject_class = (GObjectClass *) klass;
|
|
gstelement_class = (GstElementClass *) klass;
|
|
|
|
gobject_class->set_property = gst_audio_dynamic_set_property;
|
|
gobject_class->get_property = gst_audio_dynamic_get_property;
|
|
|
|
g_object_class_install_property (gobject_class, PROP_CHARACTERISTICS,
|
|
g_param_spec_enum ("characteristics", "Characteristics",
|
|
"Selects whether the ratio should be applied smooth (soft-knee) "
|
|
"or hard (hard-knee).",
|
|
GST_TYPE_AUDIO_DYNAMIC_CHARACTERISTICS, CHARACTERISTICS_HARD_KNEE,
|
|
G_PARAM_READWRITE | G_PARAM_STATIC_STRINGS));
|
|
|
|
g_object_class_install_property (gobject_class, PROP_MODE,
|
|
g_param_spec_enum ("mode", "Mode",
|
|
"Selects whether the filter should work on loud samples (compressor) or"
|
|
"quiet samples (expander).",
|
|
GST_TYPE_AUDIO_DYNAMIC_MODE, MODE_COMPRESSOR,
|
|
G_PARAM_READWRITE | G_PARAM_STATIC_STRINGS));
|
|
|
|
g_object_class_install_property (gobject_class, PROP_THRESHOLD,
|
|
g_param_spec_float ("threshold", "Threshold",
|
|
"Threshold until the filter is activated", 0.0, 1.0,
|
|
0.0,
|
|
G_PARAM_READWRITE | GST_PARAM_CONTROLLABLE | G_PARAM_STATIC_STRINGS));
|
|
|
|
g_object_class_install_property (gobject_class, PROP_RATIO,
|
|
g_param_spec_float ("ratio", "Ratio",
|
|
"Ratio that should be applied", 0.0, G_MAXFLOAT,
|
|
1.0,
|
|
G_PARAM_READWRITE | GST_PARAM_CONTROLLABLE | G_PARAM_STATIC_STRINGS));
|
|
|
|
gst_element_class_set_static_metadata (gstelement_class,
|
|
"Dynamic range controller", "Filter/Effect/Audio",
|
|
"Compressor and Expander", "Sebastian Dröge <slomo@circular-chaos.org>");
|
|
|
|
caps = gst_caps_from_string (ALLOWED_CAPS);
|
|
gst_audio_filter_class_add_pad_templates (GST_AUDIO_FILTER_CLASS (klass),
|
|
caps);
|
|
gst_caps_unref (caps);
|
|
|
|
GST_AUDIO_FILTER_CLASS (klass)->setup =
|
|
GST_DEBUG_FUNCPTR (gst_audio_dynamic_setup);
|
|
|
|
GST_BASE_TRANSFORM_CLASS (klass)->transform_ip =
|
|
GST_DEBUG_FUNCPTR (gst_audio_dynamic_transform_ip);
|
|
GST_BASE_TRANSFORM_CLASS (klass)->transform_ip_on_passthrough = FALSE;
|
|
|
|
gst_type_mark_as_plugin_api (GST_TYPE_AUDIO_DYNAMIC_CHARACTERISTICS, 0);
|
|
gst_type_mark_as_plugin_api (GST_TYPE_AUDIO_DYNAMIC_MODE, 0);
|
|
}
|
|
|
|
static void
|
|
gst_audio_dynamic_init (GstAudioDynamic * filter)
|
|
{
|
|
filter->ratio = 1.0;
|
|
filter->threshold = 0.0;
|
|
filter->characteristics = CHARACTERISTICS_HARD_KNEE;
|
|
filter->mode = MODE_COMPRESSOR;
|
|
gst_base_transform_set_in_place (GST_BASE_TRANSFORM (filter), TRUE);
|
|
gst_base_transform_set_gap_aware (GST_BASE_TRANSFORM (filter), TRUE);
|
|
}
|
|
|
|
static void
|
|
gst_audio_dynamic_set_property (GObject * object, guint prop_id,
|
|
const GValue * value, GParamSpec * pspec)
|
|
{
|
|
GstAudioDynamic *filter = GST_AUDIO_DYNAMIC (object);
|
|
|
|
switch (prop_id) {
|
|
case PROP_CHARACTERISTICS:
|
|
filter->characteristics = g_value_get_enum (value);
|
|
gst_audio_dynamic_set_process_function (filter,
|
|
GST_AUDIO_FILTER_INFO (filter));
|
|
break;
|
|
case PROP_MODE:
|
|
filter->mode = g_value_get_enum (value);
|
|
gst_audio_dynamic_set_process_function (filter,
|
|
GST_AUDIO_FILTER_INFO (filter));
|
|
break;
|
|
case PROP_THRESHOLD:
|
|
filter->threshold = g_value_get_float (value);
|
|
break;
|
|
case PROP_RATIO:
|
|
filter->ratio = g_value_get_float (value);
|
|
break;
|
|
default:
|
|
G_OBJECT_WARN_INVALID_PROPERTY_ID (object, prop_id, pspec);
|
|
break;
|
|
}
|
|
}
|
|
|
|
static void
|
|
gst_audio_dynamic_get_property (GObject * object, guint prop_id,
|
|
GValue * value, GParamSpec * pspec)
|
|
{
|
|
GstAudioDynamic *filter = GST_AUDIO_DYNAMIC (object);
|
|
|
|
switch (prop_id) {
|
|
case PROP_CHARACTERISTICS:
|
|
g_value_set_enum (value, filter->characteristics);
|
|
break;
|
|
case PROP_MODE:
|
|
g_value_set_enum (value, filter->mode);
|
|
break;
|
|
case PROP_THRESHOLD:
|
|
g_value_set_float (value, filter->threshold);
|
|
break;
|
|
case PROP_RATIO:
|
|
g_value_set_float (value, filter->ratio);
|
|
break;
|
|
default:
|
|
G_OBJECT_WARN_INVALID_PROPERTY_ID (object, prop_id, pspec);
|
|
break;
|
|
}
|
|
}
|
|
|
|
/* GstAudioFilter vmethod implementations */
|
|
|
|
static gboolean
|
|
gst_audio_dynamic_setup (GstAudioFilter * base, const GstAudioInfo * info)
|
|
{
|
|
GstAudioDynamic *filter = GST_AUDIO_DYNAMIC (base);
|
|
|
|
gst_audio_dynamic_set_process_function (filter, info);
|
|
return TRUE;
|
|
}
|
|
|
|
static void
|
|
gst_audio_dynamic_transform_hard_knee_compressor_int (GstAudioDynamic * filter,
|
|
gint16 * data, guint num_samples)
|
|
{
|
|
glong val;
|
|
glong thr_p = filter->threshold * G_MAXINT16;
|
|
glong thr_n = filter->threshold * G_MININT16;
|
|
|
|
/* Nothing to do for us if ratio is 1.0 or if the threshold
|
|
* equals 1.0. */
|
|
if (filter->threshold == 1.0 || filter->ratio == 1.0)
|
|
return;
|
|
|
|
for (; num_samples; num_samples--) {
|
|
val = *data;
|
|
|
|
if (val > thr_p) {
|
|
val = thr_p + (val - thr_p) * filter->ratio;
|
|
} else if (val < thr_n) {
|
|
val = thr_n + (val - thr_n) * filter->ratio;
|
|
}
|
|
*data++ = (gint16) CLAMP (val, G_MININT16, G_MAXINT16);
|
|
}
|
|
}
|
|
|
|
static void
|
|
gst_audio_dynamic_transform_hard_knee_compressor_float (GstAudioDynamic *
|
|
filter, gfloat * data, guint num_samples)
|
|
{
|
|
gdouble val, threshold = filter->threshold;
|
|
|
|
/* Nothing to do for us if ratio == 1.0.
|
|
* As float values can be above 1.0 we have to do something
|
|
* if threshold is greater than 1.0. */
|
|
if (filter->ratio == 1.0)
|
|
return;
|
|
|
|
for (; num_samples; num_samples--) {
|
|
val = *data;
|
|
|
|
if (val > threshold) {
|
|
val = threshold + (val - threshold) * filter->ratio;
|
|
} else if (val < -threshold) {
|
|
val = -threshold + (val + threshold) * filter->ratio;
|
|
}
|
|
*data++ = (gfloat) val;
|
|
}
|
|
}
|
|
|
|
static void
|
|
gst_audio_dynamic_transform_soft_knee_compressor_int (GstAudioDynamic * filter,
|
|
gint16 * data, guint num_samples)
|
|
{
|
|
glong val;
|
|
glong thr_p = filter->threshold * G_MAXINT16;
|
|
glong thr_n = filter->threshold * G_MININT16;
|
|
gdouble a_p, b_p, c_p;
|
|
gdouble a_n, b_n, c_n;
|
|
|
|
/* Nothing to do for us if ratio is 1.0 or if the threshold
|
|
* equals 1.0. */
|
|
if (filter->threshold == 1.0 || filter->ratio == 1.0)
|
|
return;
|
|
|
|
/* We build a 2nd degree polynomial here for
|
|
* values greater than threshold or small than
|
|
* -threshold with:
|
|
* f(t) = t, f'(t) = 1, f'(m) = r
|
|
* =>
|
|
* a = (1-r)/(2*(t-m))
|
|
* b = (r*t - m)/(t-m)
|
|
* c = t * (1 - b - a*t)
|
|
* f(x) = ax^2 + bx + c
|
|
*/
|
|
|
|
/* shouldn't happen because this would only be the case
|
|
* for threshold == 1.0 which we catch above */
|
|
g_assert (thr_p - G_MAXINT16 != 0);
|
|
g_assert (thr_n - G_MININT != 0);
|
|
|
|
a_p = (1 - filter->ratio) / (2 * (thr_p - G_MAXINT16));
|
|
b_p = (filter->ratio * thr_p - G_MAXINT16) / (thr_p - G_MAXINT16);
|
|
c_p = thr_p * (1 - b_p - a_p * thr_p);
|
|
a_n = (1 - filter->ratio) / (2 * (thr_n - G_MININT16));
|
|
b_n = (filter->ratio * thr_n - G_MININT16) / (thr_n - G_MININT16);
|
|
c_n = thr_n * (1 - b_n - a_n * thr_n);
|
|
|
|
for (; num_samples; num_samples--) {
|
|
val = *data;
|
|
|
|
if (val > thr_p) {
|
|
val = a_p * val * val + b_p * val + c_p;
|
|
} else if (val < thr_n) {
|
|
val = a_n * val * val + b_n * val + c_n;
|
|
}
|
|
*data++ = (gint16) CLAMP (val, G_MININT16, G_MAXINT16);
|
|
}
|
|
}
|
|
|
|
static void
|
|
gst_audio_dynamic_transform_soft_knee_compressor_float (GstAudioDynamic *
|
|
filter, gfloat * data, guint num_samples)
|
|
{
|
|
gdouble val;
|
|
gdouble threshold = filter->threshold;
|
|
gdouble a_p, b_p, c_p;
|
|
gdouble a_n, b_n, c_n;
|
|
|
|
/* Nothing to do for us if ratio == 1.0.
|
|
* As float values can be above 1.0 we have to do something
|
|
* if threshold is greater than 1.0. */
|
|
if (filter->ratio == 1.0)
|
|
return;
|
|
|
|
/* We build a 2nd degree polynomial here for
|
|
* values greater than threshold or small than
|
|
* -threshold with:
|
|
* f(t) = t, f'(t) = 1, f'(m) = r
|
|
* =>
|
|
* a = (1-r)/(2*(t-m))
|
|
* b = (r*t - m)/(t-m)
|
|
* c = t * (1 - b - a*t)
|
|
* f(x) = ax^2 + bx + c
|
|
*/
|
|
|
|
/* FIXME: If threshold is the same as the maximum
|
|
* we need to raise it a bit to prevent
|
|
* division by zero. */
|
|
if (threshold == 1.0)
|
|
threshold = 1.0 + 0.00001;
|
|
|
|
a_p = (1.0 - filter->ratio) / (2.0 * (threshold - 1.0));
|
|
b_p = (filter->ratio * threshold - 1.0) / (threshold - 1.0);
|
|
c_p = threshold * (1.0 - b_p - a_p * threshold);
|
|
a_n = (1.0 - filter->ratio) / (2.0 * (-threshold + 1.0));
|
|
b_n = (-filter->ratio * threshold + 1.0) / (-threshold + 1.0);
|
|
c_n = -threshold * (1.0 - b_n + a_n * threshold);
|
|
|
|
for (; num_samples; num_samples--) {
|
|
val = *data;
|
|
|
|
if (val > 1.0) {
|
|
val = 1.0 + (val - 1.0) * filter->ratio;
|
|
} else if (val > threshold) {
|
|
val = a_p * val * val + b_p * val + c_p;
|
|
} else if (val < -1.0) {
|
|
val = -1.0 + (val + 1.0) * filter->ratio;
|
|
} else if (val < -threshold) {
|
|
val = a_n * val * val + b_n * val + c_n;
|
|
}
|
|
*data++ = (gfloat) val;
|
|
}
|
|
}
|
|
|
|
static void
|
|
gst_audio_dynamic_transform_hard_knee_expander_int (GstAudioDynamic * filter,
|
|
gint16 * data, guint num_samples)
|
|
{
|
|
glong val;
|
|
glong thr_p = filter->threshold * G_MAXINT16;
|
|
glong thr_n = filter->threshold * G_MININT16;
|
|
gdouble zero_p, zero_n;
|
|
|
|
/* Nothing to do for us here if threshold equals 0.0
|
|
* or ratio equals 1.0 */
|
|
if (filter->threshold == 0.0 || filter->ratio == 1.0)
|
|
return;
|
|
|
|
/* zero crossing of our function */
|
|
if (filter->ratio != 0.0) {
|
|
zero_p = thr_p - thr_p / filter->ratio;
|
|
zero_n = thr_n - thr_n / filter->ratio;
|
|
} else {
|
|
zero_p = zero_n = 0.0;
|
|
}
|
|
|
|
if (zero_p < 0.0)
|
|
zero_p = 0.0;
|
|
if (zero_n > 0.0)
|
|
zero_n = 0.0;
|
|
|
|
for (; num_samples; num_samples--) {
|
|
val = *data;
|
|
|
|
if (val < thr_p && val > zero_p) {
|
|
val = filter->ratio * val + thr_p * (1 - filter->ratio);
|
|
} else if ((val <= zero_p && val > 0) || (val >= zero_n && val < 0)) {
|
|
val = 0;
|
|
} else if (val > thr_n && val < zero_n) {
|
|
val = filter->ratio * val + thr_n * (1 - filter->ratio);
|
|
}
|
|
*data++ = (gint16) CLAMP (val, G_MININT16, G_MAXINT16);
|
|
}
|
|
}
|
|
|
|
static void
|
|
gst_audio_dynamic_transform_hard_knee_expander_float (GstAudioDynamic * filter,
|
|
gfloat * data, guint num_samples)
|
|
{
|
|
gdouble val, threshold = filter->threshold, zero;
|
|
|
|
/* Nothing to do for us here if threshold equals 0.0
|
|
* or ratio equals 1.0 */
|
|
if (filter->threshold == 0.0 || filter->ratio == 1.0)
|
|
return;
|
|
|
|
/* zero crossing of our function */
|
|
if (filter->ratio != 0.0)
|
|
zero = threshold - threshold / filter->ratio;
|
|
else
|
|
zero = 0.0;
|
|
|
|
if (zero < 0.0)
|
|
zero = 0.0;
|
|
|
|
for (; num_samples; num_samples--) {
|
|
val = *data;
|
|
|
|
if (val < threshold && val > zero) {
|
|
val = filter->ratio * val + threshold * (1.0 - filter->ratio);
|
|
} else if ((val <= zero && val > 0.0) || (val >= -zero && val < 0.0)) {
|
|
val = 0.0;
|
|
} else if (val > -threshold && val < -zero) {
|
|
val = filter->ratio * val - threshold * (1.0 - filter->ratio);
|
|
}
|
|
*data++ = (gfloat) val;
|
|
}
|
|
}
|
|
|
|
static void
|
|
gst_audio_dynamic_transform_soft_knee_expander_int (GstAudioDynamic * filter,
|
|
gint16 * data, guint num_samples)
|
|
{
|
|
glong val;
|
|
glong thr_p = filter->threshold * G_MAXINT16;
|
|
glong thr_n = filter->threshold * G_MININT16;
|
|
gdouble zero_p, zero_n;
|
|
gdouble a_p, b_p, c_p;
|
|
gdouble a_n, b_n, c_n;
|
|
gdouble r2;
|
|
|
|
/* Nothing to do for us here if threshold equals 0.0
|
|
* or ratio equals 1.0 */
|
|
if (filter->threshold == 0.0 || filter->ratio == 1.0)
|
|
return;
|
|
|
|
/* zero crossing of our function */
|
|
zero_p = (thr_p * (filter->ratio - 1.0)) / (1.0 + filter->ratio);
|
|
zero_n = (thr_n * (filter->ratio - 1.0)) / (1.0 + filter->ratio);
|
|
|
|
if (zero_p < 0.0)
|
|
zero_p = 0.0;
|
|
if (zero_n > 0.0)
|
|
zero_n = 0.0;
|
|
|
|
/* shouldn't happen as this would only happen
|
|
* with threshold == 0.0 */
|
|
g_assert (thr_p != 0);
|
|
g_assert (thr_n != 0);
|
|
|
|
/* We build a 2n degree polynomial here for values between
|
|
* 0 and threshold or 0 and -threshold with:
|
|
* f(t) = t, f'(t) = 1, f(z) = 0, f'(z) = r
|
|
* z between 0 and t
|
|
* =>
|
|
* a = (1 - r^2) / (4 * t)
|
|
* b = (1 + r^2) / 2
|
|
* c = t * (1.0 - b - a*t)
|
|
* f(x) = ax^2 + bx + c */
|
|
r2 = filter->ratio * filter->ratio;
|
|
a_p = (1.0 - r2) / (4.0 * thr_p);
|
|
b_p = (1.0 + r2) / 2.0;
|
|
c_p = thr_p * (1.0 - b_p - a_p * thr_p);
|
|
a_n = (1.0 - r2) / (4.0 * thr_n);
|
|
b_n = (1.0 + r2) / 2.0;
|
|
c_n = thr_n * (1.0 - b_n - a_n * thr_n);
|
|
|
|
for (; num_samples; num_samples--) {
|
|
val = *data;
|
|
|
|
if (val < thr_p && val > zero_p) {
|
|
val = a_p * val * val + b_p * val + c_p;
|
|
} else if ((val <= zero_p && val > 0) || (val >= zero_n && val < 0)) {
|
|
val = 0;
|
|
} else if (val > thr_n && val < zero_n) {
|
|
val = a_n * val * val + b_n * val + c_n;
|
|
}
|
|
*data++ = (gint16) CLAMP (val, G_MININT16, G_MAXINT16);
|
|
}
|
|
}
|
|
|
|
static void
|
|
gst_audio_dynamic_transform_soft_knee_expander_float (GstAudioDynamic * filter,
|
|
gfloat * data, guint num_samples)
|
|
{
|
|
gdouble val;
|
|
gdouble threshold = filter->threshold;
|
|
gdouble zero;
|
|
gdouble a_p, b_p, c_p;
|
|
gdouble a_n, b_n, c_n;
|
|
gdouble r2;
|
|
|
|
/* Nothing to do for us here if threshold equals 0.0
|
|
* or ratio equals 1.0 */
|
|
if (filter->threshold == 0.0 || filter->ratio == 1.0)
|
|
return;
|
|
|
|
/* zero crossing of our function */
|
|
zero = (threshold * (filter->ratio - 1.0)) / (1.0 + filter->ratio);
|
|
|
|
if (zero < 0.0)
|
|
zero = 0.0;
|
|
|
|
/* shouldn't happen as this only happens with
|
|
* threshold == 0.0 */
|
|
g_assert (threshold != 0.0);
|
|
|
|
/* We build a 2n degree polynomial here for values between
|
|
* 0 and threshold or 0 and -threshold with:
|
|
* f(t) = t, f'(t) = 1, f(z) = 0, f'(z) = r
|
|
* z between 0 and t
|
|
* =>
|
|
* a = (1 - r^2) / (4 * t)
|
|
* b = (1 + r^2) / 2
|
|
* c = t * (1.0 - b - a*t)
|
|
* f(x) = ax^2 + bx + c */
|
|
r2 = filter->ratio * filter->ratio;
|
|
a_p = (1.0 - r2) / (4.0 * threshold);
|
|
b_p = (1.0 + r2) / 2.0;
|
|
c_p = threshold * (1.0 - b_p - a_p * threshold);
|
|
a_n = (1.0 - r2) / (-4.0 * threshold);
|
|
b_n = (1.0 + r2) / 2.0;
|
|
c_n = -threshold * (1.0 - b_n + a_n * threshold);
|
|
|
|
for (; num_samples; num_samples--) {
|
|
val = *data;
|
|
|
|
if (val < threshold && val > zero) {
|
|
val = a_p * val * val + b_p * val + c_p;
|
|
} else if ((val <= zero && val > 0.0) || (val >= -zero && val < 0.0)) {
|
|
val = 0.0;
|
|
} else if (val > -threshold && val < -zero) {
|
|
val = a_n * val * val + b_n * val + c_n;
|
|
}
|
|
*data++ = (gfloat) val;
|
|
}
|
|
}
|
|
|
|
/* GstBaseTransform vmethod implementations */
|
|
static GstFlowReturn
|
|
gst_audio_dynamic_transform_ip (GstBaseTransform * base, GstBuffer * buf)
|
|
{
|
|
GstAudioDynamic *filter = GST_AUDIO_DYNAMIC (base);
|
|
guint num_samples;
|
|
GstClockTime timestamp, stream_time;
|
|
GstMapInfo map;
|
|
|
|
timestamp = GST_BUFFER_TIMESTAMP (buf);
|
|
stream_time =
|
|
gst_segment_to_stream_time (&base->segment, GST_FORMAT_TIME, timestamp);
|
|
|
|
GST_DEBUG_OBJECT (filter, "sync to %" GST_TIME_FORMAT,
|
|
GST_TIME_ARGS (timestamp));
|
|
|
|
if (GST_CLOCK_TIME_IS_VALID (stream_time))
|
|
gst_object_sync_values (GST_OBJECT (filter), stream_time);
|
|
|
|
if (G_UNLIKELY (GST_BUFFER_FLAG_IS_SET (buf, GST_BUFFER_FLAG_GAP)))
|
|
return GST_FLOW_OK;
|
|
|
|
gst_buffer_map (buf, &map, GST_MAP_READWRITE);
|
|
num_samples = map.size / GST_AUDIO_FILTER_BPS (filter);
|
|
|
|
filter->process (filter, map.data, num_samples);
|
|
|
|
gst_buffer_unmap (buf, &map);
|
|
|
|
return GST_FLOW_OK;
|
|
}
|