Skip to content

Commit

Permalink
Merge branch 'swesterfeld-blepsynth-sample-accurate', closes #27
Browse files Browse the repository at this point in the history 
* swesterfeld-blepsynth-sample-accurate:
  DEVICES: blepsynth: treat negative time offsets as time offset 0 events
  DEVICES: blepsynth: remove debugging code
  DEVICES: blepsynth: change defaults for 2nd oscillator / mix
  DEVICES: blepsynth: remove event offset debug checks
  DEVICES: blepsynth/blepsynth.cc: improve updates for some parameters
  DEVICES: blepsynth/blepsynth.cc: use new filter version from dsp-research
	 - use better tanh() approximation in filter saturation function
  DEVICES: blepsynth/blepsynth.cc: set more parameter used flags
  DEVICES: blepsynth/blepsynth.cc: use switch for parameter updates
  DEVICES: blepsynth/blepsynth.cc: implement sample accurate event processing

Signed-off-by: Tim Janik <[email protected]>
  • Loading branch information
@tim-janik
tim-janik committed Nov 18, 2023
2 parents be92820 + d0b3afb commit 80927e0
Show file tree
Hide file tree
Showing 3 changed files with 168 additions and 100 deletions.
191 changes: 115 additions & 76 deletions devices/blepsynth/blepsynth.cc
View file
Original file line number Diff line number Diff line change
Expand Up @@ -348,14 +348,11 @@ class BlepSynth : public AudioProcessor {
};

enum { FILTER_TYPE_BYPASS, FILTER_TYPE_LADDER, FILTER_TYPE_SKFILTER };
int filter_type_ = 0;

static constexpr int CUTOFF_MIN_MIDI = 15;
static constexpr int CUTOFF_MAX_MIDI = 144;

bool need_update_volume_envelope_;

bool need_update_filter_envelope_;

class Voice
{
public:
Expand Down Expand Up @@ -412,15 +409,19 @@ class BlepSynth : public AudioProcessor {
const uint I = oscnum + 1;
const uint O = oscnum * (OSC2_SHAPE - OSC1_SHAPE);
const String o = string_format ("osc_%u_", I);

const double shape_default = oscnum ? -100 : 0;
const double octave_default = oscnum;

pmap.group = _("Oscillator %u", I);
pmap[O+OSC1_SHAPE] = Param { o+"shape", _("Osc %u Shape", I), _("Shp%u", I), 0, "%", { -100, 100, }, };
pmap[O+OSC1_SHAPE] = Param { o+"shape", _("Osc %u Shape", I), _("Shp%u", I), shape_default, "%", { -100, 100, }, };
pmap[O+OSC1_PULSE_WIDTH] = Param { o+"pulse_width", _("Osc %u Pulse Width", I), _("PW%u", I), 50, "%", { 0, 100, }, };
pmap[O+OSC1_SUB] = Param { o+"subharmonic", _("Osc %u Subharmonic", I), _("Sub%u", I), 0, "%", { 0, 100, }, };
pmap[O+OSC1_SUB_WIDTH] = Param { o+"subharmonic_width", _("Osc %u Subharmonic Width", I), _("SbW%u", I), 50, "%", { 0, 100, }, };
pmap[O+OSC1_SYNC] = Param { o+"sync_slave", _("Osc %u Sync Slave", I), _("Syn%u", I), 0, "Semitones", { 0, 60, }, };

pmap[O+OSC1_PITCH] = Param { o+"pitch", _("Osc %u Pitch", I), _("Pit%u", I), 0, "semitones", { -7, 7, }, };
pmap[O+OSC1_OCTAVE] = Param { o+"octave", _("Osc %u Octave", I), _("Oct%u", I), 0, "octaves", { -2, 3, }, };
pmap[O+OSC1_PITCH] = Param { o+"pitch", _("Osc %u Pitch", I), _("Pit%u", I), 0, "semitones", { -7, 7, }, };
pmap[O+OSC1_OCTAVE] = Param { o+"octave", _("Osc %u Octave", I), _("Oct%u", I), octave_default, "octaves", { -2, 3, }, };

/* TODO: unison_voices property should have stepping set to 1 */
pmap[O+OSC1_UNISON_VOICES] = Param { o+"unison_voices", _("Osc %u Unison Voices", I), _("Voi%u", I), 1, "Voices", { 1, 16, }, };
Expand All @@ -431,7 +432,7 @@ class BlepSynth : public AudioProcessor {
oscparams (0);

pmap.group = _("Mix");
pmap[MIX] = Param { "mix", _("Mix"), _("Mix"), 0, "%", { 0, 100 }, };
pmap[MIX] = Param { "mix", _("Mix"), _("Mix"), 30, "%", { 0, 100 }, };
pmap[VEL_TRACK] = Param { "vel_track", _("Velocity Tracking"), _("VelTr"), 50, "%", { 0, 100, }, };
// TODO: post_gain probably should default to 0dB once we have track/mixer volumes
pmap[POST_GAIN] = Param { "post_gain", _("Post Gain"), _("Gain"), -12, "dB", { -24, 24, }, };
Expand Down Expand Up @@ -579,30 +580,61 @@ class BlepSynth : public AudioProcessor {
void
adjust_param (uint32_t tag) override
{
if (tag == FILTER_TYPE)
switch (tag)
{
for (Voice *voice : active_voices_)
case FILTER_TYPE:
{
voice->ladder_filter_.reset();
voice->skfilter_.reset();
int new_filter_type = irintf (get_param (FILTER_TYPE));
if (new_filter_type != filter_type_)
{
filter_type_ = new_filter_type;
for (Voice *voice : active_voices_)
{
if (filter_type_ == FILTER_TYPE_LADDER)
voice->ladder_filter_.reset();
if (filter_type_ == FILTER_TYPE_SKFILTER)
voice->skfilter_.reset();
}
}
set_parameter_used (LADDER_MODE, filter_type_ == FILTER_TYPE_LADDER);
set_parameter_used (SKFILTER_MODE, filter_type_ == FILTER_TYPE_SKFILTER);
}
}
if (tag == ATTACK || tag == DECAY || tag == SUSTAIN || tag == RELEASE ||
tag == ATTACK_SLOPE || tag == DECAY_SLOPE || tag == RELEASE_SLOPE)
{
need_update_volume_envelope_ = true;
}
if (tag == FIL_ATTACK || tag == FIL_DECAY || tag == FIL_SUSTAIN || tag == FIL_RELEASE)
{
need_update_filter_envelope_ = true;
}
if (tag == VE_MODEL)
{
bool ve_has_slope = irintf (get_param (VE_MODEL)) > 0; // exponential envelope has no slope parameters
break;
case ATTACK:
case DECAY:
case SUSTAIN:
case RELEASE:
case ATTACK_SLOPE:
case DECAY_SLOPE:
case RELEASE_SLOPE:
{
for (Voice *voice : active_voices_)
update_volume_envelope (voice);
break;
}
case FIL_ATTACK:
case FIL_DECAY:
case FIL_SUSTAIN:
case FIL_RELEASE:
{
for (Voice *voice : active_voices_)
update_filter_envelope (voice);
break;
}
case VE_MODEL:
{
bool ve_has_slope = irintf (get_param (VE_MODEL)) > 0; // exponential envelope has no slope parameters

set_parameter_used (ATTACK_SLOPE, ve_has_slope);
set_parameter_used (DECAY_SLOPE, ve_has_slope);
set_parameter_used (RELEASE_SLOPE, ve_has_slope);
set_parameter_used (ATTACK_SLOPE, ve_has_slope);
set_parameter_used (DECAY_SLOPE, ve_has_slope);
set_parameter_used (RELEASE_SLOPE, ve_has_slope);
break;
}
case KEY_C: check_note (KEY_C, old_c_, 60); break;
case KEY_D: check_note (KEY_D, old_d_, 62); break;
case KEY_E: check_note (KEY_E, old_e_, 64); break;
case KEY_F: check_note (KEY_F, old_f_, 65); break;
case KEY_G: check_note (KEY_G, old_g_, 67); break;
}
}
void
Expand All @@ -622,6 +654,9 @@ class BlepSynth : public AudioProcessor {
int unison_voices = irintf (get_param (O+OSC1_UNISON_VOICES));
unison_voices = CLAMP (unison_voices, 1, 16);
osc.set_unison (unison_voices, get_param (O+OSC1_UNISON_DETUNE), get_param (O+OSC1_UNISON_STEREO) * 0.01);

set_parameter_used (O + OSC1_UNISON_DETUNE, unison_voices > 1);
set_parameter_used (O + OSC1_UNISON_STEREO, unison_voices > 1);
}
static double
perc_to_s (double perc)
Expand Down Expand Up @@ -824,9 +859,6 @@ class BlepSynth : public AudioProcessor {

auto filter_process_block = [&] (auto& filter)
{
if (need_update_filter_envelope_)
update_filter_envelope (voice);

auto gen_filter_input = [&] (float *freq_in, float *reso_in, float *drive_in, uint n_frames)
{
voice->fil_envelope_.process (freq_in, n_frames);
Expand Down Expand Up @@ -864,13 +896,12 @@ class BlepSynth : public AudioProcessor {
}
};

int filter_type = irintf (get_param (FILTER_TYPE));
if (filter_type == FILTER_TYPE_LADDER)
if (filter_type_ == FILTER_TYPE_LADDER)
{
voice->ladder_filter_.set_mode (LadderVCF::Mode (irintf (get_param (LADDER_MODE))));
filter_process_block (voice->ladder_filter_);
}
else if (filter_type == FILTER_TYPE_SKFILTER)
else if (filter_type_ == FILTER_TYPE_SKFILTER)
{
voice->skfilter_.set_mode (SKFilter::Mode (irintf (get_param (SKFILTER_MODE))));
filter_process_block (voice->skfilter_);
Expand Down Expand Up @@ -902,54 +933,21 @@ class BlepSynth : public AudioProcessor {
voice->fil_envelope_.set_release (perc_to_s (get_param (FIL_RELEASE)));
}
void
render (uint n_frames) override
render_audio (float *left_out, float *right_out, uint n_frames)
{
MidiEventInput evinput = midi_event_input();
for (const auto &ev : evinput)
switch (ev.message())
{
case MidiMessage::NOTE_OFF:
note_off (ev.channel, ev.key);
break;
case MidiMessage::NOTE_ON:
note_on (ev.channel, ev.key, ev.velocity);
break;
case MidiMessage::ALL_NOTES_OFF:
for (auto voice : active_voices_)
if (voice->state_ == Voice::ON && voice->channel_ == ev.channel)
note_off (voice->channel_, voice->midi_note_);
break;
case MidiMessage::PARAM_VALUE:
apply_event (ev);
adjust_param (ev.param);
break;
default: ;
}

/* TODO: replace this with true midi input */
check_note (KEY_C, old_c_, 60);
check_note (KEY_D, old_d_, 62);
check_note (KEY_E, old_e_, 64);
check_note (KEY_F, old_f_, 65);
check_note (KEY_G, old_g_, 67);
if (!n_frames)
return;

assert_return (n_ochannels (stereout_) == 2);
bool need_free = false;
float *left_out = oblock (stereout_, 0);
float *right_out = oblock (stereout_, 1);

floatfill (left_out, 0.f, n_frames);
floatfill (right_out, 0.f, n_frames);

for (Voice *voice : active_voices_)
{
if (voice->new_voice_)
{
int filter_type = irintf (get_param (FILTER_TYPE));
int idelay = 0;
if (filter_type == FILTER_TYPE_LADDER)
if (filter_type_ == FILTER_TYPE_LADDER)
idelay = voice->ladder_filter_.delay();
if (filter_type == FILTER_TYPE_SKFILTER)
if (filter_type_ == FILTER_TYPE_SKFILTER)
idelay = voice->skfilter_.delay();
if (idelay)
{
Expand All @@ -966,8 +964,6 @@ class BlepSynth : public AudioProcessor {

// apply volume envelope
float volume_env[n_frames];
if (need_update_volume_envelope_)
update_volume_envelope (voice);
voice->envelope_.process (volume_env, n_frames);
float post_gain_factor = db2voltage (get_param (POST_GAIN));
for (uint i = 0; i < n_frames; i++)
Expand All @@ -984,9 +980,52 @@ class BlepSynth : public AudioProcessor {
}
if (need_free)
free_unused_voices();
need_update_volume_envelope_ = false;
need_update_filter_envelope_ = false;
}
void
render (uint n_frames) override
{
assert_return (n_ochannels (stereout_) == 2);

float *left_out = oblock (stereout_, 0);
float *right_out = oblock (stereout_, 1);

floatfill (left_out, 0.f, n_frames);
floatfill (right_out, 0.f, n_frames);

uint offset = 0;
MidiEventInput evinput = midi_event_input();
for (const auto &ev : evinput)
{
uint frame = std::max<int> (ev.frame, 0); // TODO: should be unsigned anyway, issue #26

// process any audio that is before the event
render_audio (left_out + offset, right_out + offset, frame - offset);
offset = frame;

switch (ev.message())
{
case MidiMessage::NOTE_OFF:
note_off (ev.channel, ev.key);
break;
case MidiMessage::NOTE_ON:
note_on (ev.channel, ev.key, ev.velocity);
break;
case MidiMessage::ALL_NOTES_OFF:
for (auto voice : active_voices_)
if (voice->state_ == Voice::ON && voice->channel_ == ev.channel)
note_off (voice->channel_, voice->midi_note_);
break;
case MidiMessage::PARAM_VALUE:
apply_event (ev);
adjust_param (ev.param);
break;
default: ;
}
}
// process frames after last event
render_audio (left_out + offset, right_out + offset, n_frames - offset);
}

static double
convert_cutoff (double midi_note)
{
Expand Down
37 changes: 26 additions & 11 deletions devices/blepsynth/laddervcf.hh
View file
Original file line number Diff line number Diff line change
Expand Up @@ -6,6 +6,9 @@
#include "pandaresampler.hh"

#include <array>
#include <algorithm>
#include <cassert>
#include <cmath>

namespace Ase {

Expand Down Expand Up @@ -36,6 +39,7 @@ private:
float freq_ = 440;
float reso_ = 0;
float drive_ = 0;
float global_volume_ = 1;
uint over_ = 0;
bool test_linear_ = false;

Expand Down Expand Up @@ -86,6 +90,17 @@ public:
fparams_valid_ = false;
}
void
set_global_volume (float global_volume)
{
/* every samples that is processed by the filter is
* - multiplied with global_volume before processing
* - divided by global_volume after processing
* which has an effect on the non-linear part of the filter (drive)
*/
global_volume_ = global_volume;
fparams_valid_ = false;
}
void
set_test_linear (bool test_linear)
{
test_linear_ = test_linear;
Expand Down Expand Up @@ -135,14 +150,6 @@ private:
clamp_freq_min_ = frequency_range_min_;
clamp_freq_max_ = std::min (frequency_range_max_, rate_ * over_ * 0.49f);
}
float
distort (float x)
{
/* shaped somewhat similar to tanh() and others, but faster */
x = std::clamp (x, -1.0f, 1.0f);

return x - x * x * x * (1.0f / 3);
}
void
setup_reso_drive (FParams& fparams, float reso, float drive)
{
Expand Down Expand Up @@ -171,10 +178,18 @@ private:
reso += drive * sqrt (reso) * reso * 0.03f;

float vol = exp2f ((drive + -12 * sqrt (reso)) * db_x2_factor);
fparams.pre_scale = negative_drive_vol * vol;
fparams.post_scale = std::max (1 / vol, 1.0f);
fparams.pre_scale = negative_drive_vol * vol * global_volume_;
fparams.post_scale = std::max (1 / vol, 1.0f) / global_volume_;
fparams.reso = sqrt (reso) * 4;
}
static float
tanh_approx (float x)
{
// https://www.musicdsp.org/en/latest/Other/238-rational-tanh-approximation.html
x = std::clamp (x, -3.0f, 3.0f);

return x * (27.0f + x * x) / (27.0f + 9.0f * x * x);
}
/*
* This ladder filter implementation is mainly based on
*
Expand Down Expand Up @@ -203,7 +218,7 @@ private:
Channel& c = channels_[i];
const float x = value * fparams_.pre_scale;
const float g_comp = 0.5f; // passband gain correction
const float x0 = distort (x - (c.y4 - g_comp * x) * res);
const float x0 = tanh_approx (x - (c.y4 - g_comp * x) * res);

c.y1 = b0 * x0 + b1 * c.x1 - a1 * c.y1;
c.x1 = x0;
Expand Down
Loading

0 comments on commit 80927e0

Please sign in to comment.