From 63a00e17442ad02ffb9bff89440bdf494ccc7cf1 Mon Sep 17 00:00:00 2001
From: Chris <chris@ternar.tech>
Date: Tue, 21 Oct 2025 11:06:00 +0200
Subject: [PATCH] add smoothing to split eq mode transitions

---
 filter/spliteq.h  | 180 ++++++++++++++++++++++++++++------------------
 util/audio_math.h |   2 +
 util/smoother.h   |  60 ++++++++++++++++
 3 files changed, 174 insertions(+), 68 deletions(-)
 create mode 100644 util/smoother.h

diff --git a/filter/spliteq.h b/filter/spliteq.h
index b3803f6..4a8475e 100644
--- a/filter/spliteq.h
+++ b/filter/spliteq.h
@@ -1,5 +1,6 @@
 #pragma once
 #include "audio_math.h"
+#include "smoother.h"
 #include <cmath>
 #include <vector>
 
@@ -224,6 +225,11 @@ inline void aw_filter_process_block(aw_filter& f, float* audio, int frames)
 	}
 }
 
+enum spliteq_mode {
+	CASCADE_SUM,
+	LINKWITZ_RILEY
+};
+
 struct spliteq {
 	aw_filter lp_l, lp_r, hp_l, hp_r; // lowpass and highpass filters
 
@@ -257,7 +263,10 @@ struct spliteq {
 	float mid_gain_adj = 1.0f;
 	float treble_gain_adj = 1.0f;
 
-	bool linkwitz_riley_enabled = false;
+	spliteq_mode current_mode = LINKWITZ_RILEY;
+	spliteq_mode target_mode = LINKWITZ_RILEY;
+	bool transitioning = false;
+	smoother transition_smoother;
 };
 
 inline void spliteq_init(spliteq& eq, double samplerate, double low_mid_crossover, double mid_high_crossover)
@@ -367,86 +376,121 @@ inline void spliteq_init(spliteq& eq, double samplerate, double low_mid_crossove
 	eq.treble2_r.cutoff = mid_high_crossover;
 	eq.treble2_r.x1 = eq.treble2_r.x2 = eq.treble2_r.y1 = eq.treble2_r.y2 = 0.0;
 	butterworth_biquad_coeffs(eq.treble2_r, samplerate);
+
+	smoother_init(eq.transition_smoother, samplerate, 50.0f, 1.0f);
+}
+
+inline void spliteq_set_mode(spliteq& eq, spliteq_mode mode)
+{
+	if (eq.target_mode != mode) {
+		eq.target_mode = mode;
+		eq.transitioning = true;
+		smoother_set_target(eq.transition_smoother, 0.0f);
+	}
+}
+
+inline void linkwitz_riley_process(spliteq& eq, float**& audio, int& i)
+{
+	// left channel
+	double input_l = audio[0][i];
+	// bass
+	double bass_l = butterworth_biquad_process(eq.bass1_l, input_l);
+	bass_l = butterworth_biquad_process(eq.bass2_l, bass_l);
+	// mid
+	double mid_l = butterworth_biquad_process(eq.mid_hp1_l, input_l);
+	mid_l = butterworth_biquad_process(eq.mid_hp2_l, mid_l);
+	mid_l = butterworth_biquad_process(eq.mid_lp1_l, mid_l);
+	mid_l = butterworth_biquad_process(eq.mid_lp2_l, mid_l);
+	// treble
+	double treble_l = butterworth_biquad_process(eq.treble1_l, input_l);
+	treble_l = butterworth_biquad_process(eq.treble2_l, treble_l);
+
+	// apply gains
+	bass_l *= eq.bass_gain;
+	mid_l *= eq.mid_gain;
+	treble_l *= eq.treble_gain;
+
+	// sum bands
+	audio[0][i] = bass_l + mid_l + treble_l;
+
+	// right channel
+	double input_r = audio[1][i];
+	double bass_r = butterworth_biquad_process(eq.bass1_r, input_r);
+	bass_r = butterworth_biquad_process(eq.bass2_r, bass_r);
+
+	double mid_r = butterworth_biquad_process(eq.mid_hp1_r, input_r);
+	mid_r = butterworth_biquad_process(eq.mid_hp2_r, mid_r);
+	mid_r = butterworth_biquad_process(eq.mid_lp1_r, mid_r);
+	mid_r = butterworth_biquad_process(eq.mid_lp2_r, mid_r);
+
+	double treble_r = butterworth_biquad_process(eq.treble1_r, input_r);
+	treble_r = butterworth_biquad_process(eq.treble2_r, treble_r);
+
+	bass_r *= eq.bass_gain;
+	mid_r *= eq.mid_gain;
+	treble_r *= eq.treble_gain;
+
+	audio[1][i] = bass_r + mid_r + treble_r;
+}
+
+inline void cascade_sum_process(spliteq& eq, float**& audio, int& i)
+{
+	double input_l = audio[0][i];
+	double input_r = audio[1][i];
+
+	double bass_l = cascade_filter_process(eq.bass_l, input_l);
+	double bass_r = cascade_filter_process(eq.bass_r, input_r);
+
+	double treble_l = cascade_filter_process(eq.treble_l, input_l);
+	double treble_r = cascade_filter_process(eq.treble_r, input_r);
+
+	double mid_l = input_l - bass_l - treble_l;
+	double mid_r = input_r - bass_r - treble_r;
+
+	// apply gains
+	bass_l *= eq.bass_gain_adj;
+	bass_r *= eq.bass_gain_adj;
+	mid_l *= eq.mid_gain_adj;
+	mid_r *= eq.mid_gain_adj;
+	treble_l *= eq.treble_gain_adj;
+	treble_r *= eq.treble_gain_adj;
+
+	// sum bands
+	audio[0][i] = bass_l + mid_l + treble_l;
+	audio[1][i] = bass_r + mid_r + treble_r;
 }
 
-// Process block (stereo)
 inline void spliteq_process_block(spliteq& eq, float** audio, int frames)
 {
+	// highpass filters
 	aw_filter_process_block(eq.hp_l, audio[0], frames);
 	aw_filter_process_block(eq.hp_r, audio[1], frames);
 
 	for (int i = 0; i < frames; i++) {
-		if (eq.linkwitz_riley_enabled) {
-			// butterwort/linkwitz-riley
+		float smooth_gain = 1.0f;
 
-			// Left channel
-			double input_l = audio[0][i];
-			// Bass
-			double bass_l = butterworth_biquad_process(eq.bass1_l, input_l);
-			bass_l = butterworth_biquad_process(eq.bass2_l, bass_l);
-			// Mid
-			double mid_l = butterworth_biquad_process(eq.mid_hp1_l, input_l);
-			mid_l = butterworth_biquad_process(eq.mid_hp2_l, mid_l);
-			mid_l = butterworth_biquad_process(eq.mid_lp1_l, mid_l);
-			mid_l = butterworth_biquad_process(eq.mid_lp2_l, mid_l);
-			// Treble
-			double treble_l = butterworth_biquad_process(eq.treble1_l, input_l);
-			treble_l = butterworth_biquad_process(eq.treble2_l, treble_l);
+		if (eq.transitioning) {
+			smooth_gain = smoother_process_sample(eq.transition_smoother);
 
-			// Apply gains
-			bass_l *= eq.bass_gain;
-			mid_l *= eq.mid_gain;
-			treble_l *= eq.treble_gain;
-
-			// Sum bands
-			audio[0][i] = bass_l + mid_l + treble_l;
-
-			// Right channel
-			double input_r = audio[1][i];
-			double bass_r = butterworth_biquad_process(eq.bass1_r, input_r);
-			bass_r = butterworth_biquad_process(eq.bass2_r, bass_r);
-
-			double mid_r = butterworth_biquad_process(eq.mid_hp1_r, input_r);
-			mid_r = butterworth_biquad_process(eq.mid_hp2_r, mid_r);
-			mid_r = butterworth_biquad_process(eq.mid_lp1_r, mid_r);
-			mid_r = butterworth_biquad_process(eq.mid_lp2_r, mid_r);
-
-			double treble_r = butterworth_biquad_process(eq.treble1_r, input_r);
-			treble_r = butterworth_biquad_process(eq.treble2_r, treble_r);
-
-			bass_r *= eq.bass_gain;
-			mid_r *= eq.mid_gain;
-			treble_r *= eq.treble_gain;
-
-			audio[1][i] = bass_r + mid_r + treble_r;
-		} else {
-			// cascade/sum
-			double input_l = audio[0][i];
-			double input_r = audio[1][i];
-
-			double bass_l = cascade_filter_process(eq.bass_l, input_l);
-			double bass_r = cascade_filter_process(eq.bass_r, input_r);
-
-			double treble_l = cascade_filter_process(eq.treble_l, input_l);
-			double treble_r = cascade_filter_process(eq.treble_r, input_r);
-
-			double mid_l = input_l - bass_l - treble_l;
-			double mid_r = input_r - bass_r - treble_r;
-
-			// Apply gains
-			bass_l *= eq.bass_gain_adj;
-			bass_r *= eq.bass_gain_adj;
-			mid_l *= eq.mid_gain_adj;
-			mid_r *= eq.mid_gain_adj;
-			treble_l *= eq.treble_gain_adj;
-			treble_r *= eq.treble_gain_adj;
-
-			// Sum bands
-			audio[0][i] = bass_l + mid_l + treble_l;
-			audio[1][i] = bass_r + mid_r + treble_r;
+			if (smooth_gain == 0.f) {
+				smoother_set_target(eq.transition_smoother, 1.0);
+				eq.current_mode = eq.target_mode;
+			} else if (smooth_gain == 1.f) {
+				eq.transitioning = false;
+			}
 		}
+
+		if (eq.current_mode == LINKWITZ_RILEY) {
+			linkwitz_riley_process(eq, audio, i);
+		} else if (eq.current_mode == CASCADE_SUM) {
+			cascade_sum_process(eq, audio, i);
+		}
+
+		audio[0][i] *= smooth_gain;
+		audio[1][i] *= smooth_gain;
 	}
 
+	// lowpass filters
 	aw_filter_process_block(eq.lp_l, audio[0], frames);
 	aw_filter_process_block(eq.lp_r, audio[1], frames);
 }
diff --git a/util/audio_math.h b/util/audio_math.h
index 30a2b73..d99bd85 100644
--- a/util/audio_math.h
+++ b/util/audio_math.h
@@ -12,4 +12,6 @@ inline double lin_2_db(double lin)
 inline double db_2_lin(double db) { return pow(10.0, db / 20.0); }
 
 inline float midi_to_frequency(float midi_note) { return 440.0 * powf(2.0, ((float)midi_note - 69.0) / 12.0); }
+
+inline float ms_to_samples(float ms, double sample_rate) { return (ms * 0.001f) * (float)sample_rate; }
 } // namespace trnr
\ No newline at end of file
diff --git a/util/smoother.h b/util/smoother.h
new file mode 100644
index 0000000..b0fcedd
--- /dev/null
+++ b/util/smoother.h
@@ -0,0 +1,60 @@
+#pragma once
+
+#include "audio_math.h"
+
+namespace trnr {
+
+struct smoother {
+	float samplerate;
+	float time_samples;
+
+	float current;
+	float target;
+	float increment;
+	int32_t remaining;
+};
+
+inline void smoother_init(smoother& s, double samplerate, float time_ms, float initial_value = 0.0f)
+{
+	s.samplerate = fmax(0.0, samplerate);
+	s.time_samples = ms_to_samples(time_ms, s.samplerate);
+	s.current = initial_value;
+	s.target = initial_value;
+	s.increment = 0.0f;
+	s.remaining = 0;
+}
+
+inline void smoother_set_target(smoother& s, float newTarget)
+{
+	s.target = newTarget;
+
+	// immediate if time is zero or too short
+	if (s.time_samples <= 1.0f) {
+		s.current = s.target;
+		s.increment = 0.0f;
+		s.remaining = 0;
+		return;
+	}
+
+	int32_t n = static_cast<int32_t>(fmax(1.0f, ceilf(s.time_samples)));
+	s.remaining = n;
+	// protect against denormals / tiny differences, but we want exact reach at the end
+	s.increment = (s.target - s.current) / static_cast<float>(s.remaining);
+}
+
+// process a single sample and return the smoothed value
+inline float smoother_process_sample(smoother& s)
+{
+	if (s.remaining > 0) {
+		s.current += s.increment;
+		--s.remaining;
+		if (s.remaining == 0) {
+			// ensure exact target at the end to avoid FP drift
+			s.current = s.target;
+			s.increment = 0.0f;
+		}
+	}
+	return s.current;
+}
+
+} // namespace trnr
\ No newline at end of file