synth-post-pass.cpp

/*
	FM. BISON hybrid FM synthesis -- Post-processing pass.
	(C) njdewit technologies (visualizers.nl) & bipolaraudio.nl
	MIT license applies, please see https://en.wikipedia.org/wiki/MIT_License or LICENSE in the project root!

	- Copy
	- Auto-wah/Vox
	- Yamaha Reface CP-style chorus & phaser
	- Delay
	- Tube distortion (4X oversampling)
	- Post filter (24dB) (4X oversampling)
	- Reverb
	- Compressor
	- Low cut, 3-band tuning, master volume & final clamp

	This grew into a huge function; that was of course not the intention but so far it's functional and quite well
	documented so right now (01/07/2020) I see no reason to chop it up
*/

#include "synth-post-pass.h"
#include "synth-stateless-oscillators.h"
#include "synth-distort.h"
#include "patch/synth-patch-global.h"

namespace SFM
{
	// Remedies sampling artifacts whilst sweeping a delay line
	constexpr float kSweepCutoffHz = 50.f;

	// Max. delay feedback (so as not to create an endless loop)
	constexpr float kMaxDelayFeedback = 0.95f; // Like Ableton does, or so I've been told by Paul

	// Delay line size (main delay) & cross bleed amount
	constexpr float kMainDelayLineSize = kMainDelayInSec;
	constexpr float kDelayCrossbleeding = kGoldenRatio*0.1f; // Arbitrary
	
	// The compressor's 'bite' is filtered so it can be used as a GUI indicator; higher value means brighter and quicker
	constexpr float kCompressorBiteCutHz = 480.f;

	// "Tape delay" constants (for 'wow' effect)
	constexpr float kTapeDelayHz = kGoldenRatio;
	constexpr float kTapeDelaySpread = 0.02f;

	// Tube tone LPF Qs (in normalized domain, [0...1])
	constexpr float kTubeToneFlatQ = 0.f;
	constexpr float kTubeToneColorQ = kGoldenRatio*0.0628f;

	PostPass::PostPass(unsigned sampleRate, unsigned maxSamplesPerBlock, unsigned Nyquist) :
		m_sampleRate(sampleRate), m_Nyquist(Nyquist), m_sampleRate4X(sampleRate*4)

		// Delay
,		m_tapeDelayLFO(sampleRate)
,		m_tapeDelayLPF(kSweepCutoffHz /* Borrowed, FIXME */ / sampleRate)
,		m_delayLineL(unsigned(sampleRate*kMainDelayLineSize))
,		m_delayLineM(unsigned(sampleRate*kMainDelayLineSize))
,		m_delayLineR(unsigned(sampleRate*kMainDelayLineSize))
,		m_curDelayInSec(0.f, sampleRate, kDefParameterLatency * 4.f /* Longer */, 0.f, kMainDelayInSec)
,		m_curDelayWet(0.f, sampleRate, kDefParameterLatency, 0.f, 1.f)
,		m_curDelayDrive(1.f /* 0dB */, sampleRate, kDefParameterLatency, 0.f, 1.f)
,		m_curDelayFeedback(0.f, sampleRate, kDefParameterLatency, 0.f, 1.f)
,		m_curDelayFeedbackCutoff(1.f, sampleRate, kDefParameterLatency, 0.f, 1.f)
,		m_curDelayTapeWow(0.f, sampleRate, kDefParameterLatency, 0.f, 1.f)
		
		// Chorus/Phaser
,		m_chorusDL(sampleRate/10  /* 100MS max. chorus delay */)
,		m_chorusSweep(sampleRate), m_chorusSweepMod(sampleRate)
,		m_chorusSweepLPF1(kSweepCutoffHz/sampleRate), m_chorusSweepLPF2(kSweepCutoffHz/sampleRate)
,		m_phaserSweep(sampleRate)
,		m_phaserSweepLPF((kSweepCutoffHz*2.f)/sampleRate) // Tweaked a little for effect

		// Oversampling (stereo)
,		m_oversampling4X(2, 2, juce::dsp::Oversampling<float>::filterHalfBandFIREquiripple, true)

		// Post filter
,		m_postFilter(m_sampleRate4X)
,		m_curPostCutoff(0.f, m_sampleRate4X, kDefParameterLatency * 2.f /* Longer */, 0.f, 1.f)
,		m_curPostReso(0.f, m_sampleRate4X, kDefParameterLatency, 0.f, 1.f)
,		m_curPostDrive(0.f, m_sampleRate4X, kDefParameterLatency, 0.f, 1.f)
,		m_curPostWet(0.f, m_sampleRate4X, kDefParameterLatency, 0.f, 1.f)
		
		// Tube distort
,		m_curTubeDist(0.f, m_sampleRate4X, kDefParameterLatency, 0.f, 1.f)
,		m_curTubeDrive(kDefTubeDrive, m_sampleRate4X, kDefParameterLatency, 0.f, 1.f)
,		m_curTubeOffset(0.f, m_sampleRate4X, kDefParameterLatency, 0.f, 1.f)
,		m_curTubeTone(kDefTubeTone, m_sampleRate4X, kDefParameterLatency, 0.f, 1.f)

		// Post (EQ)
,		m_postEQ(sampleRate, true)

		// External effects
,		m_wah(sampleRate, Nyquist)
,		m_reverb(sampleRate, Nyquist)
,		m_compressor(sampleRate)
,		m_compressorBiteLPF(kCompressorBiteCutHz/sampleRate)
		
		// Misc.
,		m_curChorusWet(0.f, sampleRate, kDefParameterLatency, 0.f, 1.f)
,		m_curPhaserWet(0.f, sampleRate, kDefParameterLatency, 0.f, 1.f)
,		m_curMasterVol(1.f, sampleRate, kDefParameterLatency, 0.f, 1.f)
	{
		// Allocate intermediate buffers
		m_pBufL  = reinterpret_cast<float *>(mallocAligned(maxSamplesPerBlock*sizeof(float), 16));
		m_pBufR  = reinterpret_cast<float *>(mallocAligned(maxSamplesPerBlock*sizeof(float), 16));

		// Initialize JUCE oversampling object (FIXME)
		m_oversampling4X.initProcessing(maxSamplesPerBlock);

		// Set tape delay mod. frequency
		m_tapeDelayLFO.Initialize(kTapeDelayHz, m_sampleRate);

		// Set low kill (cut) filter
		m_killLow.reset();
		m_killLow.setBiquad(bq_type_highpass, kLowCutHz/sampleRate, kLowCutQ, 0.f);
	}

	PostPass::~PostPass()
	{
		freeAligned(m_pBufL);
		freeAligned(m_pBufR);
	}

	float PostPass::GetLatency() const
	{
		// FIXME: approx. complete sum best possible
		const float oversamplingLatency = m_oversampling4X.getLatencyInSamples();
		const float compressorLatency   = m_compressor.GetLatency();

		return oversamplingLatency + compressorLatency;
	}

	void PostPass::Apply(unsigned numSamples,
	                     float rateBPM, unsigned overideFlagsRateBPM,
	                     float wahResonance, float wahAttack, float wahHold, float wahRate, float wahDrivedB, float wahSpeak, float wahSpeakVowel, float wahSpeakVowelMod, float wahSpeakGhost, float wahSpeakCut, float wahSpeakReso, float wahCut, float wahWet,
	                     float cpRate, float cpWet, bool isChorus,
	                     float delayInSec, float delayWet, float delayDrivedB, float delayFeedback, float delayFeedbackCutoff, float delayTapeWow,
	                     float postCutoff, float postReso, float postDrivedB, float postWet,
	                     float tubeDistort, float tubeDrive, float tubeOffset, float tubeTone, bool tubeToneReso,
	                     float reverbWet, float reverbRoomSize, float reverbDampening, float reverbWidth, float reverbLP, float reverbHP, float reverbPreDelay,
	                     float compThresholddB, float compKneedB, float compRatio, float compGaindB, float compAttack, float compRelease, float compLookahead, bool compAutoGain, float compRMSToPeak,
	                     float bassTuningdB, float trebleTuningdB, float midTuningdB, float masterVoldB,
	                     const float *pLeftIn, const float *pRightIn, float *pLeftOut, float *pRightOut)
	{
		// Shitload of assertions; some values are asserted in functions they're passed to (make this a habit) plus this might not be 100% complete (FIXME)
		SFM_ASSERT(nullptr != pLeftIn  && nullptr != pRightIn);
		SFM_ASSERT(nullptr != pLeftOut && nullptr != pRightOut);
		SFM_ASSERT(numSamples > 0);
		SFM_ASSERT(rateBPM >= 0.f);
		SFM_ASSERT_NORM(cpRate);
		SFM_ASSERT_NORM(cpWet);
		SFM_ASSERT(delayInSec >= 0.f && delayInSec <= kMainDelayInSec);
		SFM_ASSERT_NORM(delayWet);
		SFM_ASSERT(delayDrivedB >= kMinDelayDrivedB && delayDrivedB <= kMaxDelayDrivedB);
		SFM_ASSERT_NORM(delayFeedback);
		SFM_ASSERT_NORM(delayTapeWow);
		SFM_ASSERT_NORM(delayFeedbackCutoff);
		SFM_ASSERT_NORM(postCutoff);
		SFM_ASSERT_NORM(postReso);
		SFM_ASSERT(masterVoldB >= kMinVolumedB && masterVoldB <= kMaxVolumedB);
		SFM_ASSERT_NORM(tubeDistort);
		SFM_ASSERT(tubeDrive >= kMinTubeDrive && tubeDrive <= kMaxTubeDrive);
		SFM_ASSERT(tubeOffset >= kMinTubeOffset && tubeOffset <= kMaxTubeOffset);
		SFM_ASSERT_NORM(tubeTone);
		
		// Delay is automatically overridden to it's manual setting if it doesn't fit in it's delay line
		const bool useBPM = 0.f != rateBPM;
		
		// BPM sync. overrides (LFO is handled in FM_BISON.cpp)
		const bool overrideSyncAW    = overideFlagsRateBPM & kFlagOverrideAW;
		const bool overrideSyncCP    = overideFlagsRateBPM & kFlagOverrideCP;
		const bool overrideSyncDelay = overideFlagsRateBPM & kFlagOverrideDelay;

		/* ----------------------------------------------------------------------------------------------------

			Copy samples to local buffers

		 ------------------------------------------------------------------------------------------------------ */

		const size_t bufSize = numSamples * sizeof(float);

		memcpy(m_pBufL, pLeftIn,  bufSize);
		memcpy(m_pBufR, pRightIn, bufSize);

#if !defined(SFM_DISABLE_FX)

		/* ----------------------------------------------------------------------------------------------------

			Auto-wah w/'Vox'

		 ------------------------------------------------------------------------------------------------------ */

		if (true == useBPM && false == overrideSyncAW)
			wahRate = rateBPM; // Tested, works fine!

		m_wah.SetParameters(wahResonance, wahAttack, wahHold, wahRate, wahDrivedB, wahSpeak, wahSpeakVowel, wahSpeakVowelMod, wahSpeakGhost, wahSpeakCut, wahSpeakReso, wahCut, wahWet);
		m_wah.Apply(m_pBufL, m_pBufR, numSamples, false == useBPM);

		/* ----------------------------------------------------------------------------------------------------

			Chorus/Phaser + Delay

			Mixed on top of original signal, does not introduce latency.

		 ------------------------------------------------------------------------------------------------------ */

		if (true == isChorus)
		{
			m_curChorusWet.SetTarget(cpWet);
			m_curPhaserWet.SetTarget(0.f);
		}
		else
		{
			m_curChorusWet.SetTarget(0.f);
			m_curPhaserWet.SetTarget(cpWet);
		}
		
		// Calculate delay & set delay mode
		const float delay = (false == useBPM || true == overrideSyncDelay) ? delayInSec : 1.f/rateBPM;
		SFM_ASSERT(delay >= 0.f && delay <= kMainDelayInSec);

		// Set delay param. targets
		m_curDelayInSec.SetTarget(delay);
		m_curDelayWet.SetTarget(delayWet);
		m_curDelayDrive.SetTarget(dB2Lin(delayDrivedB));
		m_curDelayFeedback.SetTarget(delayFeedback);
		m_curDelayFeedbackCutoff.SetTarget(delayFeedbackCutoff);
		m_curDelayTapeWow.SetTarget(delayTapeWow);
		
		// Set rate for both chorus & phaser
		if (false == useBPM || true == overrideSyncCP) // Sync. to BPM?
		{
			// No, use manual setting
			SetChorusRate(cpRate, kMaxChorusRate);
			SetPhaserRate(cpRate, kMaxPhaserRate);
		}
		else
		{
			// Locked to BPM
			static_assert(kMaxChorusRate >= kMaxPhaserRate);
			SetChorusRate(rateBPM, kMaxChorusRate/kMaxPhaserRate);
			SetPhaserRate(rateBPM, 1.f);
		}
				
		for (unsigned iSample = 0; iSample < numSamples; ++iSample)
		{
			const float sampleL = m_pBufL[iSample];
			const float sampleR = m_pBufR[iSample];

			float left = sampleL, right = sampleR;
				
			// Always feed the chorus delay line
			// This approach has it's flaws: https://matthewvaneerde.wordpress.com/2010/12/07/downmixing-stereo-to-mono/
			m_chorusDL.Write(left*0.5f + right*0.5f);
			
			//
			// Apply chorus & phaser
			//

			const float chorusWet = m_curChorusWet.Sample();
			const float phaserWet = m_curPhaserWet.Sample();

			// Breaking my own 'execute the entire chain' rule here
			if (chorusWet > 0.f) 
				ApplyChorus(sampleL, sampleR, left, right, chorusWet);
			
			if (phaserWet > 0.f)
				ApplyPhaser(left, right, left, right, phaserWet);

			//
			// Apply delay (always executed, for now, FIXME?)
			//

			const float monaural = left*0.5f + right*0.5f;

			const float curDelayInSec = m_curDelayInSec.Sample();
			SFM_ASSERT(curDelayInSec >= 0.f && curDelayInSec <= kMainDelayInSec);

			// Write driven samples to delay line
			const float drive = m_curDelayDrive.Sample();
			m_delayLineL.Write(left     * drive);
			m_delayLineM.Write(monaural * drive);
			m_delayLineR.Write(right    * drive);
			
			// Sample delay line
			const float curDelay   = curDelayInSec/kMainDelayInSec;
			const float curTapeWow = m_curDelayTapeWow.Sample();
			const float normDelay  = curDelay + curTapeWow*(curDelay*curDelay)*kTapeDelaySpread*m_tapeDelayLPF.Apply(fast_cosf(m_tapeDelayLFO.Sample()));
			const float delayedL   = m_delayLineL.ReadNormalized(normDelay);
			const float delayedM   = m_delayLineM.ReadNormalized(normDelay);
			const float delayedR   = m_delayLineR.ReadNormalized(normDelay);
			
			// Bleed delay samples a bit
			constexpr float crossBleedAmt = kDelayCrossbleeding;
			constexpr float invCrossBleedAmt = 1.f-crossBleedAmt;
			const float crossBleed = delayedM;
			const float delayL = delayedL*invCrossBleedAmt + crossBleed*crossBleedAmt;
			const float delayR = delayedR*invCrossBleedAmt + crossBleed*crossBleedAmt;

			// Filter delay
			const float curFc = (m_curDelayFeedbackCutoff.Sample() * m_Nyquist/4)/m_sampleRate; // Limited range gives a more pronounced effect
			m_delayFeedbackLPF_L.SetCutoff(curFc);
			m_delayFeedbackLPF_R.SetCutoff(curFc);

			const float filteredL = m_delayFeedbackLPF_L.Apply(delayL);
			const float filteredR = m_delayFeedbackLPF_R.Apply(delayR);

			const float filteredM = 0.5f*filteredL + 0.5f*filteredR;

			// Feedback
			const float curFeedback =  m_curDelayFeedback.Sample()*kMaxDelayFeedback;
			m_delayLineL.WriteFeedback(filteredL, curFeedback);
			m_delayLineM.WriteFeedback(filteredM, curFeedback);
			m_delayLineR.WriteFeedback(filteredR, curFeedback);

			// Add delay

//			const float wet = m_curDelayWet.Sample();
//			m_pBufL[iSample] = left  + wet*delayL;
//			m_pBufR[iSample] = right + wet*delayR;

			// Stereo (width) effect (fixed)
			// Nicked from synth-reverb.cpp
			const float wet = m_curDelayWet.Sample();
			const float dry = 1.f-wet;

			const float width = kGoldenRatio; // FIXME: parameter?
			const float wet1  = wet*(width*0.5f + 0.5f);
			const float wet2  = wet*((1.f-width)*0.5f);
			
//			m_pBufL[iSample] = delayL*wet1 + delayR*wet2 + left*dry;
//			m_pBufR[iSample] = delayR*wet1 + delayL*wet2 + right*dry;
			
			// To be more like Ableton, we'll use the filtered samples rightaway
			m_pBufL[iSample] = filteredL*wet1 + filteredR*wet2 + left*dry;
			m_pBufR[iSample] = filteredR*wet1 + filteredL*wet2 + right*dry;
		}

		/* ----------------------------------------------------------------------------------------------------

			Oversampled: 24dB ladder filter & tube distortion (4X)

			JUCE says:
			" Choose between FIR or IIR filtering depending on your needs in term of latency and phase 
			  distortion. With FIR filters, the phase is linear but the latency is maximised. With IIR 
			  filtering, the phase is compromised around the Nyquist frequency but the latency is minimised. "
		 
		 	Currently using FIR (ergo low phase distortion over latency).
		 
		 	Not oversampling degrades the quality of the effects; the distortion is much more prone to coarse
		 	transients and aliasing and the stability of the 24dB filter kernel is less.

		 ------------------------------------------------------------------------------------------------------ */
                        
		// Set post filter parameters
		m_curPostCutoff.SetTarget(postCutoff);
		m_curPostReso.SetTarget(postReso);
		m_curPostDrive.SetTarget(dBToGain(postDrivedB));
		m_curPostWet.SetTarget(postWet);

		// Set tube distortion parameters
		m_curTubeDist.SetTarget(tubeDistort);
		m_curTubeDrive.SetTarget(tubeDrive);
		m_curTubeOffset.SetTarget(tubeOffset);
		m_curTubeTone.SetTarget(tubeTone);

		const float toneQ = SVF_ResoToQ(tubeToneReso ? kTubeToneColorQ : kTubeToneFlatQ);
		
		// Main buffers
		float *inputBuffers[2] = { m_pBufL, m_pBufR };
		juce::dsp::AudioBlock<float> inputBlock(inputBuffers, 2, numSamples);

		// Oversample 4X
		auto outBlock = m_oversampling4X.processSamplesUp(inputBlock);
		const size_t numOversamples = outBlock.getNumSamples();
		SFM_ASSERT(numOversamples == numSamples*4);

		float *pOverL = outBlock.getChannelPointer(0);
		float *pOverR = outBlock.getChannelPointer(1);

		for (unsigned iSample = 0; iSample < numOversamples; ++iSample)
		{
			float sampleL = pOverL[iSample]; 
			float sampleR = pOverR[iSample];

			// Apply (non-linear) distortion
			const float amount = m_curTubeDist.Sample();
			const float drive  = m_curTubeDrive.Sample();
			const float offset = m_curTubeOffset.Sample();
			const float tone   = m_curTubeTone.Sample();

			// Apply (soft) clipping
			float inDistSampleL = sampleL, inDistSampleR = sampleR;

			// Apply tone filter (resonant LPF)
			m_tubeToneFilter.updateLowpassCoeff(SVF_CutoffToHz(tone, m_Nyquist), toneQ, m_sampleRate4X);
			m_tubeToneFilter.tick(inDistSampleL, inDistSampleR);

//			const float driveAdj = drive/kMaxTubeDrive; // Normalized
//			float distortedL = Squarepusher(offset+sampleL, driveAdj);
//			float distortedR = Squarepusher(offset+sampleR, driveAdj);
			float distortedL = ZoelzerClip((offset+inDistSampleL)*drive); // CubicClip(offset+sampleL, drive);
			float distortedR = ZoelzerClip((offset+inDistSampleR)*drive); // CubicClip(offset+sampleR, drive);

			// Remove possible DC offset
			m_tubeDCBlocker.Apply(distortedL, distortedR);

			// Add to signal
//			float postDistortedL = sampleL + distortedL*amount; // lerpf<float>(sampleL, distortedL, smoothstepped);
//			float postDistortedR = sampleR + distortedR*amount; // lerpf<float>(sampleR, distortedR, smoothstepped);
//			const float smoothstepped = smoothstepf(amount);
			float postDistortedL = lerpf<float>(sampleL, distortedL, amount);
			float postDistortedR = lerpf<float>(sampleR, distortedR, amount);

			// Apply 24dB post filter
			const float curPostCutoff = m_curPostCutoff.Sample();
			const float curPostReso   = m_curPostReso.Sample();
			const float curPostDrive  = m_curPostDrive.Sample();
			const float curPostWet    = m_curPostWet.Sample();

			// Apply filter
			float filteredL = postDistortedL, filteredR = postDistortedR;
			m_postFilter.SetParameters(kMinPostFilterCutoffHz + curPostCutoff*kPostFilterCutoffRange, curPostReso /* [0..1] */, curPostDrive);
			m_postFilter.Apply(filteredL, filteredR);

			// Blend
			sampleL = lerpf<float>(postDistortedL, filteredL, curPostWet);
			sampleR = lerpf<float>(postDistortedR, filteredR, curPostWet);

			// Write
			pOverL[iSample] = sampleL;
			pOverR[iSample] = sampleR;
		}

		// Downsample result
		m_oversampling4X.processSamplesDown(inputBlock);

		/* ----------------------------------------------------------------------------------------------------

			Reverb

			Could possibly introduce minor latency; please check (FIXME).

		 ------------------------------------------------------------------------------------------------------ */

		// Apply reverb (after post filter to avoid muddy sound)
		m_reverb.SetRoomSize(reverbRoomSize);
		m_reverb.SetDampening(reverbDampening);
		m_reverb.SetWidth(reverbWidth);
		m_reverb.SetPreDelay(reverbPreDelay);
		m_reverb.Apply(m_pBufL, m_pBufR, numSamples, reverbWet, reverbLP, reverbHP);

		/* ----------------------------------------------------------------------------------------------------

			Compressor

			- Causes latency when 'compLookahead' is larger than zero
			- Returns 'bite', which practically means if compression has taken place

		 ------------------------------------------------------------------------------------------------------ */

		 m_compressor.SetParameters(compThresholddB, compKneedB, compRatio, compGaindB, compAttack, compRelease, compLookahead);
		 m_compressorBiteLPF.Apply(m_compressor.Apply(m_pBufL, m_pBufR, numSamples, compAutoGain, compRMSToPeak));
		 
#endif

		/* ----------------------------------------------------------------------------------------------------

			Final pass

		 ------------------------------------------------------------------------------------------------------ */
		
		// Set master volume target
		m_curMasterVol.SetTarget(dBToGain(masterVoldB));

		// Set EQ target
		m_postEQ.SetTargetdBs(bassTuningdB, trebleTuningdB, midTuningdB);

		for (unsigned iSample = 0; iSample < numSamples; ++iSample)
		{
			float sampleL = m_pBufL[iSample];
			float sampleR = m_pBufR[iSample];

			// EQ
			m_postEQ.Apply(sampleL, sampleR);

			// Apply gain (master volume)
			const float gain = m_curMasterVol.Sample();
			sampleL *= gain;
			sampleR *= gain;

			// Low cut
			m_killLow.process(sampleL, sampleR);

			// Clamp because DAWs like it like that
			pLeftOut[iSample]  = Clamp(sampleL);
			pRightOut[iSample] = Clamp(sampleR);
		}
	}

	/* ----------------------------------------------------------------------------------------------------

		Chorus/Phaser impl.

		FIXME: these are *old* and could use a review or rewrite

	 ------------------------------------------------------------------------------------------------------ */

	void PostPass::ApplyChorus(float sampleL, float sampleR, float &outL, float &outR, float wetness)
	{
		// Sweep modulation LFO
		const float sweepMod = fast_cosf(m_chorusSweepMod.Sample());
		
		// Sweep LFOs
		const float phase = m_chorusSweep.Sample();
		
		// This violates my [0..1] phase rule but Kusma's cosine approximation function handles it just fine
		const float sweepL = 0.5f*fast_sinf(phase+sweepMod);
		const float sweepR = 0.5f*fast_sinf((1.f-phase)+sweepMod);
		
		// 2 sweeping samples (FIXME: parametrize, though it's not a priority)
		const float delay  = m_sampleRate*0.005f; // 50MS delay
		const float spread = m_sampleRate*0.003f; // Sweep 30MS
		
		SFM_ASSERT(delay  < m_chorusDL.size());
		SFM_ASSERT(spread < m_chorusDL.size());
		
		// Take sweeped L/R samples (filtered to circumvent artifacts)
		const float chorusL = m_chorusDL.Read(delay + spread*m_chorusSweepLPF1.Apply(sweepL));
		const float chorusR = m_chorusDL.Read(delay + spread*m_chorusSweepLPF2.Apply(sweepR));
		
		// Add result to dry signal
		wetness *= kMaxChorusPhaserWet;
		outL = sampleL + wetness*chorusL; 
		outR = sampleR + wetness*chorusR; 
	}

	void PostPass::ApplyPhaser(float sampleL, float sampleR, float &outL, float &outR, float wetness)
	{
		// Sweep LFO (filtered for pleasing effect)
		const float sweepMod = m_phaserSweepLPF.Apply(oscTriangle(m_phaserSweep.Sample()));
		
		// Sweep cutoff frequency around center
		constexpr float range = 0.2f;
		static_assert(range < 0.5f);
		const float normCutoff = 0.5f + range*sweepMod;
		
		// Start with dry sample
		float filteredL = sampleL;
		float filteredR = sampleR;

		// Cutoff & Q
		const float cutoffHz = SVF_CutoffToHz(normCutoff, m_Nyquist);
		float Q = kSVFLowestFilterQ; // FIXME: use higher Q?
		
		// Apply cascading filters
		for (auto &filter : m_allpassFilters)
		{
			filter.updateAllpassCoeff(cutoffHz, Q, m_sampleRate);
			filter.tick(filteredL, filteredR);

			// Adds a little "space"
			Q += Q;
		}
		
		// Add result to dry signal
		wetness *= kMaxChorusPhaserWet;
		outL = sampleL + wetness*filteredL;
		outR = sampleR + wetness*filteredR;
	}
}