Collidoscope

/*

 Copyright (C) 2002 James McCartney.

 Copyright (C) 2016  Queen Mary University of London 

 Author: Fiore Martin, based on Supercollider's (http://supercollider.github.io) TGrains code and Ross Bencina's "Implementing Real-Time Granular Synthesis"

 This file is part of Collidoscope.

 Collidoscope is free software: you can redistribute it and/or modify

 it under the terms of the GNU General Public License as published by

 the Free Software Foundation, either version 3 of the License, or

 (at your option) any later version.

 This program 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 General Public License for more details.

 You should have received a copy of the GNU General Public License

 along with this program.  If not, see <http://www.gnu.org/licenses/>.

*/

#pragma once

#include <array>

#include <type_traits>

#include "EnvASR.h"

namespace collidoscope {

using std::size_t;

/**

 * The very core of the Collidoscope audio engine: the granular synthesizer.

 * Based on SuperCollider's TGrains and Ross Bencina's "Implementing Real-Time Granular Synthesis" 

 *

 * It implements Collidoscope's selection-based approach to granular synthesis. 

 * A grain is basically a selection of a recorded sample of audio. 

 * Grains are played in a loop: they are re-triggered each time they reach the end of the selection.

 * However, if the duration coefficient is greater than one, a new grain is re-triggered before the previous one is done,

 * the grains start to overlap with each other and create the typical eerie sound of grnular synthesis.

 * Also every time a new grain is triggered, it is offset of a few samples from the initial position to make the timbre more interesting.

 *

 *

 * PGranular uses a linear ASR envelope with 10 milliseconds attack and 50 milliseconds release.

 *

 * Note that PGranular is header based and only depends on std library and on "EnvASR.h" (also header based).

 * This means you can embedd it in two your project just by copying these two files over.

 *

 * Template arguments: 

 * T: type of the audio samples (normally float or double) 

 * RandOffsetFunc: type of the callable passed as argument to the contructor

 * TriggerCallbackFunc: type of the callable passed as argument to the contructor

 *

 */ 

template <typename T, typename RandOffsetFunc, typename TriggerCallbackFunc>

class PGranular

{

public:

    static const size_t kMaxGrains = 32;

    static const size_t kMinGrainsDuration = 640;

    static inline T interpolateLin( double xn, double xn_1, double decimal )

    {

        /* weighted sum interpolation */

        return static_cast<T> ((1 - decimal) * xn + decimal * xn_1);

    }

    /**

     * A single grain of the granular synthesis 

     */ 

    struct PGrain

    {

        double phase;    // read pointer to mBuffer of this grain 

        double rate;     // rate of the grain. e.g. rate = 2 the grain will play twice as fast

        bool alive;      // whether this grain is alive. Not alive means it has been processed and can be replaced by another grain

        size_t age;      // age of this grain in samples 

        size_t duration; // duration of this grain in samples. minimum = 4

        double b1;       // hann envelope from Ross Becina's "Implementing real time Granular Synthesis"

        double y1;

        double y2;

    };

    /**

     * Constructor.

     *

     * \param buffer a pointer to an array of T that contains the original sample that will be granulized

     * \param bufferLen length of buffer in samples 

     * \rand function of type size_t ()(void) that is called back each time a new grain is generated. The returned value is used 

     * to offset the starting sample of the grain. This adds more colour to the sound especially with small selections. 

     * \triggerCallback function of type void ()(char, int) that is called back each time a new grain is generated.

     *      The function is passed the character 't' as first parameter when a new grain is triggered and the characted 'e' when the synth becomes idle (no sound).

     * \ID id of this PGrain. Passed to the triggerCallback function as second parameter to identify this PGranular as the caller.

     */ 

    PGranular( const T* buffer, size_t bufferLen, size_t sampleRate, RandOffsetFunc & rand, TriggerCallbackFunc & triggerCallback, int ID ) :

        mBuffer( buffer ),

        mBufferLen( bufferLen ),

        mNumAliveGrains( 0 ),

        mGrainsRate( 1.0 ),

        mTrigger( 0 ),

        mTriggerRate( 0 ), // start silent 

        mGrainsStart( 0 ),

        mGrainsDuration( kMinGrainsDuration ),

        mGrainsDurationCoeff( 1 ),

        mRand( rand ),

        mTriggerCallback( triggerCallback ),

        mEnvASR( 1.0f, 0.01f, 0.05f, sampleRate ),

        mAttenuation( T(0.25118864315096) ),

        mID( ID )

    {

        static_assert(std::is_pod<PGrain>::value, "PGrain must be POD");

#ifdef _WINDOW

        static_assert(std::is_same<std::result_of<RandOffsetFunc()>::type, size_t>::value, "Rand must return a size_t");

#endif

        /* init the grains */

        for ( size_t grainIdx = 0; grainIdx < kMaxGrains; grainIdx++ ){

            mGrains[grainIdx].phase = 0;

            mGrains[grainIdx].rate = 1;

            mGrains[grainIdx].alive = false;

            mGrains[grainIdx].age = 0;

            mGrains[grainIdx].duration = 1;

        }

    }

    ~PGranular(){}

    /** Sets multiplier of duration of grains in seconds */

    void setGrainsDurationCoeff( double coeff )

    {

        mGrainsDurationCoeff = coeff;

        mGrainsDuration = std::lround( mTriggerRate * coeff ); 

        if ( mGrainsDuration < kMinGrainsDuration )

            mGrainsDuration = kMinGrainsDuration;

    }

    /** Sets rate of grains. e.g rate = 2 means one octave higer */

    void setGrainsRate( double rate )

    {

        mGrainsRate = rate;

    }

    /** sets the selection start in samples */

    void setSelectionStart( size_t start )

    {

        mGrainsStart = start;

    }

    /** Sets the selection size ( and therefore the trigger rate) in samples */

    void setSelectionSize( size_t size )

    {

        if ( size < kMinGrainsDuration )

            size = kMinGrainsDuration;

        mTriggerRate = size;

        mGrainsDuration = std::lround( size * mGrainsDurationCoeff );

    }

    /** Sets the attenuation of the grains with respect to the level of the recorded sample

     *  attenuation is in amp value and defaule value is 0.25118864315096 (-12dB) */

    void setAttenuation( T attenuation )

    {

        mAttenuation = attenuation;

    }

    /** Starts the synthesis engine */

    void noteOn( double rate )

    {

        if ( mEnvASR.getState() == EnvASR<T>::State::eIdle ){

            // note on sets triggering top the min value 

            if ( mTriggerRate < kMinGrainsDuration ){

                mTriggerRate = kMinGrainsDuration;

            }

            setGrainsRate( rate );

            mEnvASR.setState( EnvASR<T>::State::eAttack );

        }

    }

    /** Stops the synthesis engine */

    void noteOff()

    {

        if ( mEnvASR.getState() != EnvASR<T>::State::eIdle ){

            mEnvASR.setState( EnvASR<T>::State::eRelease );

        }

    }

    /** Whether the synthesis engine is active or not. After noteOff is called the synth stays active until the envelope decays to 0 */

    bool isIdle()

    {

        return mEnvASR.getState() == EnvASR<T>::State::eIdle;

    }

    /**

     * Runs the granular engine and stores the output in \a audioOut

     * 

     * \param pointer to an array of T. This will be filled with the output of PGranular. It needs to be at least \a numSamples long

     * \param tempBuffer a temporary buffer used to store the envelope value. It needs to be at least \a numSamples long

     * \param numSamples number of samples to be processed 

     */ 

    void process( T* audioOut, T* tempBuffer, size_t numSamples )

    {

        // num samples worth of sound ( due to envelope possibly finishing )

        size_t envSamples = 0;

        bool becameIdle = false;

        // process the envelope first and store it in the tempBuffer 

        for ( size_t i = 0; i < numSamples; i++ ){

            tempBuffer[i] = mEnvASR.tick();

            envSamples++;

            if ( isIdle() ){

                // means that the envelope has stopped 

                becameIdle = true;

                break;

            }

        }

        // does the actual grains processing 

        processGrains( audioOut, tempBuffer, envSamples );

        // becomes idle if the envelope goes to idle state 

        if ( becameIdle ){

            mTriggerCallback( 'e', mID );

            reset();

        }

    }

private:

    void processGrains( T* audioOut, T* envelopeValues, size_t numSamples )

    {

        /* process all existing alive grains */

        for ( size_t grainIdx = 0; grainIdx < mNumAliveGrains;  ){

            synthesizeGrain( mGrains[grainIdx], audioOut, envelopeValues, numSamples );

            if ( !mGrains[grainIdx].alive ){

                // this grain is dead so copy the last of the active grains here 

                // so as to keep all active grains at the beginning of the array 

                // don't increment grainIdx so the last active grain is processed next cycle

                // if this grain is the last active grain then mNumAliveGrains is decremented 

                // and grainIdx = mNumAliveGrains so the loop stops 

                copyGrain( mNumAliveGrains - 1, grainIdx );

                mNumAliveGrains--;

            }

            else{

                // go to next grain 

                grainIdx++;

            }

        }

        if ( mTriggerRate == 0 ){

            return;

        }

        size_t randOffset =  mRand();

        bool newGrainWasTriggered = false;

        // trigger new grain and synthesize them as well 

        while ( mTrigger < numSamples ){

            // if there is room to accommodate new grains 

            if ( mNumAliveGrains < kMaxGrains ){

                // get next grain will be placed at the end of the alive ones 

                size_t grainIdx = mNumAliveGrains;

                mNumAliveGrains++;

                // initialize and synthesise the grain 

                PGrain &grain = mGrains[grainIdx];

                double phase = mGrainsStart + double( randOffset );

                if ( phase >= mBufferLen )

                    phase -= mBufferLen;

                grain.phase = phase;

                grain.rate = mGrainsRate;

                grain.alive = true;

                grain.age = 0;

                grain.duration = mGrainsDuration;

                const double w = 3.14159265358979323846 / mGrainsDuration;

                grain.b1 = 2.0 * std::cos( w );

                grain.y1 = std::sin( w );

                grain.y2 = 0.0;

                synthesizeGrain( grain, audioOut + mTrigger, envelopeValues + mTrigger, numSamples - mTrigger );

                if ( grain.alive == false ) {

                    mNumAliveGrains--;

                }

                newGrainWasTriggered = true;

            }

            // update trigger even if no new grain was started 

            mTrigger += mTriggerRate;

        }

        // prepare trigger for next cycle: init mTrigger with the reminder of the samples from this cycle 

        mTrigger -= numSamples;

        if ( newGrainWasTriggered ){

            mTriggerCallback( 't', mID );

        }

    }

    // synthesize a single grain 

    // audioOut = pointer to audio block to fill 

    // numSamples = number of samples to process for this block

    void synthesizeGrain( PGrain &grain, T* audioOut, T* envelopeValues, size_t numSamples )

    {

        // copy all grain data into local variable for faster processing

        const auto rate = grain.rate;

        auto phase = grain.phase;

        auto age = grain.age;

        auto duration = grain.duration;

        auto b1 = grain.b1;

        auto y1 = grain.y1;

        auto y2 = grain.y2;

        // only process minimum between samples of this block and time left to leave for this grain 

        auto numSamplesToOut = std::min( numSamples, duration - age );

        for ( size_t sampleIdx = 0; sampleIdx < numSamplesToOut; sampleIdx++ ){

            const size_t readIndex = (size_t)phase;

            const size_t nextReadIndex = (readIndex == mBufferLen - 1) ? 0 : readIndex + 1; // wrap on the read buffer if needed 

            const double decimal = phase - readIndex;

            T out = interpolateLin( mBuffer[readIndex], mBuffer[nextReadIndex], decimal );

            // apply raised cosine bell envelope 

            auto y0 = b1 * y1 - y2;

            y2 = y1;

            y1 = y0;

            out *= T(y0);

            audioOut[sampleIdx] += out * envelopeValues[sampleIdx] * mAttenuation;

            // increment age one sample 

            age++;

            // increment the phase according to the rate of this grain 

            phase += rate;

            if ( phase >= mBufferLen ){   // wrap the phase if needed 

                phase -= mBufferLen;

            }

        }

        if ( age == duration ){

            // if it processed all the samples left to leave ( numSamplesToOut = duration-age)

            // then the grain is finished 

            grain.alive = false;

        }

        else{

            grain.phase = phase;

            grain.age = age;

            grain.y1 = y1;

            grain.y2 = y2;

        }

    }

    void copyGrain( size_t from, size_t to)

    {

        mGrains[to] = mGrains[from];

    }

    void reset()

    {

        mTrigger = 0;

        for ( size_t i = 0; i < mNumAliveGrains; i++ ){

            mGrains[i].alive = false;

        }

        mNumAliveGrains = 0;

    }

    int mID;

    // pointer to (mono) buffer, where the underlying sample is recorder 

    const T* mBuffer;

    // length of mBuffer in samples 

    const size_t mBufferLen;

    // offset in the buffer where the grains start. a.k.a. selection start 

    size_t mGrainsStart;

    // attenuates signal prevents clipping of grains (to some degree)

    T mAttenuation;

    // grain duration in samples 

    double mGrainsDurationCoeff;

    // duration of grains is selection size * duration coeff

    size_t mGrainsDuration;

    // rate of grain, affects pitch 

    double mGrainsRate;

    size_t mTrigger;       // next onset

    size_t mTriggerRate;   // inter onset

    // the array of grains 

    std::array<PGrain, kMaxGrains> mGrains;

    // number of alive grains 

    size_t mNumAliveGrains;

    RandOffsetFunc &mRand;

    TriggerCallbackFunc &mTriggerCallback;

    EnvASR<T> mEnvASR;

};

} // namespace collidoscope