alc/effects/vmorpher.cpp

   1 /**
   2  * This file is part of the OpenAL Soft cross platform audio library
   3  *
   4  * Copyright (C) 2019 by Anis A. Hireche
   5  *
   6  * Redistribution and use in source and binary forms, with or without
   7  * modification, are permitted provided that the following conditions are met:
   8  *
   9  * * Redistributions of source code must retain the above copyright notice,
  10  *   this list of conditions and the following disclaimer.
  11  *
  12  * * Redistributions in binary form must reproduce the above copyright notice,
  13  *   this list of conditions and the following disclaimer in the documentation
  14  *   and/or other materials provided with the distribution.
  15  *
  16  * * Neither the name of Spherical-Harmonic-Transform nor the names of its
  17  *   contributors may be used to endorse or promote products derived from
  18  *   this software without specific prior written permission.
  19  *
  20  * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
  21  * AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
  22  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
  23  * ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT HOLDER OR CONTRIBUTORS BE
  24  * LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
  25  * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
  26  * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
  27  * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
  28  * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
  29  * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
  30  * POSSIBILITY OF SUCH DAMAGE.
  31  */
  32
  33 #include "config.h"
  34
  35 #include <algorithm>
  36 #include <array>
  37 #include <cstdlib>
  38 #include <functional>
  39 #include <iterator>
  40
  41 #include "alc/effects/base.h"
  42 #include "almalloc.h"
  43 #include "alnumbers.h"
  44 #include "alnumeric.h"
  45 #include "alspan.h"
  46 #include "core/ambidefs.h"
  47 #include "core/bufferline.h"
  48 #include "core/context.h"
  49 #include "core/devformat.h"
  50 #include "core/device.h"
  51 #include "core/effectslot.h"
  52 #include "core/mixer.h"
  53 #include "intrusive_ptr.h"
  54
  55
  56 namespace {
  57
  58 using uint = unsigned int;
  59
  60 #define MAX_UPDATE_SAMPLES 256
  61 #define NUM_FORMANTS       4
  62 #define NUM_FILTERS        2
  63 #define Q_FACTOR           5.0f
  64
  65 #define VOWEL_A_INDEX      0
  66 #define VOWEL_B_INDEX      1
  67
  68 #define WAVEFORM_FRACBITS  24
  69 #define WAVEFORM_FRACONE   (1<<WAVEFORM_FRACBITS)
  70 #define WAVEFORM_FRACMASK  (WAVEFORM_FRACONE-1)
  71
  72 inline float Sin(uint index)
  73 {
  74     constexpr float scale{al::numbers::pi_v<float>*2.0f / WAVEFORM_FRACONE};
  75     return std::sin(static_cast<float>(index) * scale)*0.5f + 0.5f;
  76 }
  77
  78 inline float Saw(uint index)
  79 { return static_cast<float>(index) / float{WAVEFORM_FRACONE}; }
  80
  81 inline float Triangle(uint index)
  82 { return std::fabs(static_cast<float>(index)*(2.0f/WAVEFORM_FRACONE) - 1.0f); }
  83
  84 inline float Half(uint) { return 0.5f; }
  85
  86 template<float (&func)(uint)>
  87 void Oscillate(float *RESTRICT dst, uint index, const uint step, size_t todo)
  88 {
  89     for(size_t i{0u};i < todo;i++)
  90     {
  91         index += step;
  92         index &= WAVEFORM_FRACMASK;
  93         dst[i] = func(index);
  94     }
  95 }
  96
  97 struct FormantFilter
  98 {
  99     float mCoeff{0.0f};
 100     float mGain{1.0f};
 101     float mS1{0.0f};
 102     float mS2{0.0f};
 103
 104     FormantFilter() = default;
 105     FormantFilter(float f0norm, float gain)
 106       : mCoeff{std::tan(al::numbers::pi_v<float> * f0norm)}, mGain{gain}
 107     { }
 108
 109     inline void process(const float *samplesIn, float *samplesOut, const size_t numInput)
 110     {
 111         /* A state variable filter from a topology-preserving transform.
 112          * Based on a talk given by Ivan Cohen: https://www.youtube.com/watch?v=esjHXGPyrhg
 113          */
 114         const float g{mCoeff};
 115         const float gain{mGain};
 116         const float h{1.0f / (1.0f + (g/Q_FACTOR) + (g*g))};
 117         float s1{mS1};
 118         float s2{mS2};
 119
 120         for(size_t i{0u};i < numInput;i++)
 121         {
 122             const float H{(samplesIn[i] - (1.0f/Q_FACTOR + g)*s1 - s2)*h};
 123             const float B{g*H + s1};
 124             const float L{g*B + s2};
 125
 126             s1 = g*H + B;
 127             s2 = g*B + L;
 128
 129             // Apply peak and accumulate samples.
 130             samplesOut[i] += B * gain;
 131         }
 132         mS1 = s1;
 133         mS2 = s2;
 134     }
 135
 136     inline void clear()
 137     {
 138         mS1 = 0.0f;
 139         mS2 = 0.0f;
 140     }
 141 };
 142
 143
 144 struct VmorpherState final : public EffectState {
 145     struct {
 146         /* Effect parameters */
 147         FormantFilter Formants[NUM_FILTERS][NUM_FORMANTS];
 148
 149         /* Effect gains for each channel */
 150         float CurrentGains[MAX_OUTPUT_CHANNELS]{};
 151         float TargetGains[MAX_OUTPUT_CHANNELS]{};
 152     } mChans[MaxAmbiChannels];
 153
 154     void (*mGetSamples)(float*RESTRICT, uint, const uint, size_t){};
 155
 156     uint mIndex{0};
 157     uint mStep{1};
 158
 159     /* Effects buffers */
 160     alignas(16) float mSampleBufferA[MAX_UPDATE_SAMPLES]{};
 161     alignas(16) float mSampleBufferB[MAX_UPDATE_SAMPLES]{};
 162     alignas(16) float mLfo[MAX_UPDATE_SAMPLES]{};
 163
 164     void deviceUpdate(const DeviceBase *device, const Buffer &buffer) override;
 165     void update(const ContextBase *context, const EffectSlot *slot, const EffectProps *props,
 166         const EffectTarget target) override;
 167     void process(const size_t samplesToDo, const al::span<const FloatBufferLine> samplesIn,
 168         const al::span<FloatBufferLine> samplesOut) override;
 169
 170     static std::array<FormantFilter,4> getFiltersByPhoneme(VMorpherPhenome phoneme,
 171         float frequency, float pitch);
 172
 173     DEF_NEWDEL(VmorpherState)
 174 };
 175
 176 std::array<FormantFilter,4> VmorpherState::getFiltersByPhoneme(VMorpherPhenome phoneme,
 177     float frequency, float pitch)
 178 {
 179     /* Using soprano formant set of values to
 180      * better match mid-range frequency space.
 181      *
 182      * See: https://www.classes.cs.uchicago.edu/archive/1999/spring/CS295/Computing_Resources/Csound/CsManual3.48b1.HTML/Appendices/table3.html
 183      */
 184     switch(phoneme)
 185     {
 186     case VMorpherPhenome::A:
 187         return {{
 188             {( 800 * pitch) / frequency, 1.000000f}, /* std::pow(10.0f,   0 / 20.0f); */
 189             {(1150 * pitch) / frequency, 0.501187f}, /* std::pow(10.0f,  -6 / 20.0f); */
 190             {(2900 * pitch) / frequency, 0.025118f}, /* std::pow(10.0f, -32 / 20.0f); */
 191             {(3900 * pitch) / frequency, 0.100000f}  /* std::pow(10.0f, -20 / 20.0f); */
 192         }};
 193     case VMorpherPhenome::E:
 194         return {{
 195             {( 350 * pitch) / frequency, 1.000000f}, /* std::pow(10.0f,   0 / 20.0f); */
 196             {(2000 * pitch) / frequency, 0.100000f}, /* std::pow(10.0f, -20 / 20.0f); */
 197             {(2800 * pitch) / frequency, 0.177827f}, /* std::pow(10.0f, -15 / 20.0f); */
 198             {(3600 * pitch) / frequency, 0.009999f}  /* std::pow(10.0f, -40 / 20.0f); */
 199         }};
 200     case VMorpherPhenome::I:
 201         return {{
 202             {( 270 * pitch) / frequency, 1.000000f}, /* std::pow(10.0f,   0 / 20.0f); */
 203             {(2140 * pitch) / frequency, 0.251188f}, /* std::pow(10.0f, -12 / 20.0f); */
 204             {(2950 * pitch) / frequency, 0.050118f}, /* std::pow(10.0f, -26 / 20.0f); */
 205             {(3900 * pitch) / frequency, 0.050118f}  /* std::pow(10.0f, -26 / 20.0f); */
 206         }};
 207     case VMorpherPhenome::O:
 208         return {{
 209             {( 450 * pitch) / frequency, 1.000000f}, /* std::pow(10.0f,   0 / 20.0f); */
 210             {( 800 * pitch) / frequency, 0.281838f}, /* std::pow(10.0f, -11 / 20.0f); */
 211             {(2830 * pitch) / frequency, 0.079432f}, /* std::pow(10.0f, -22 / 20.0f); */
 212             {(3800 * pitch) / frequency, 0.079432f}  /* std::pow(10.0f, -22 / 20.0f); */
 213         }};
 214     case VMorpherPhenome::U:
 215         return {{
 216             {( 325 * pitch) / frequency, 1.000000f}, /* std::pow(10.0f,   0 / 20.0f); */
 217             {( 700 * pitch) / frequency, 0.158489f}, /* std::pow(10.0f, -16 / 20.0f); */
 218             {(2700 * pitch) / frequency, 0.017782f}, /* std::pow(10.0f, -35 / 20.0f); */
 219             {(3800 * pitch) / frequency, 0.009999f}  /* std::pow(10.0f, -40 / 20.0f); */
 220         }};
 221     default:
 222         break;
 223     }
 224     return {};
 225 }
 226
 227
 228 void VmorpherState::deviceUpdate(const DeviceBase*, const Buffer&)
 229 {
 230     for(auto &e : mChans)
 231     {
 232         std::for_each(std::begin(e.Formants[VOWEL_A_INDEX]), std::end(e.Formants[VOWEL_A_INDEX]),
 233             std::mem_fn(&FormantFilter::clear));
 234         std::for_each(std::begin(e.Formants[VOWEL_B_INDEX]), std::end(e.Formants[VOWEL_B_INDEX]),
 235             std::mem_fn(&FormantFilter::clear));
 236         std::fill(std::begin(e.CurrentGains), std::end(e.CurrentGains), 0.0f);
 237     }
 238 }
 239
 240 void VmorpherState::update(const ContextBase *context, const EffectSlot *slot,
 241     const EffectProps *props, const EffectTarget target)
 242 {
 243     const DeviceBase *device{context->mDevice};
 244     const float frequency{static_cast<float>(device->Frequency)};
 245     const float step{props->Vmorpher.Rate / frequency};
 246     mStep = fastf2u(clampf(step*WAVEFORM_FRACONE, 0.0f, float{WAVEFORM_FRACONE-1}));
 247
 248     if(mStep == 0)
 249         mGetSamples = Oscillate<Half>;
 250     else if(props->Vmorpher.Waveform == VMorpherWaveform::Sinusoid)
 251         mGetSamples = Oscillate<Sin>;
 252     else if(props->Vmorpher.Waveform == VMorpherWaveform::Triangle)
 253         mGetSamples = Oscillate<Triangle>;
 254     else /*if(props->Vmorpher.Waveform == VMorpherWaveform::Sawtooth)*/
 255         mGetSamples = Oscillate<Saw>;
 256
 257     const float pitchA{std::pow(2.0f,
 258         static_cast<float>(props->Vmorpher.PhonemeACoarseTuning) / 12.0f)};
 259     const float pitchB{std::pow(2.0f,
 260         static_cast<float>(props->Vmorpher.PhonemeBCoarseTuning) / 12.0f)};
 261
 262     auto vowelA = getFiltersByPhoneme(props->Vmorpher.PhonemeA, frequency, pitchA);
 263     auto vowelB = getFiltersByPhoneme(props->Vmorpher.PhonemeB, frequency, pitchB);
 264
 265     /* Copy the filter coefficients to the input channels. */
 266     for(size_t i{0u};i < slot->Wet.Buffer.size();++i)
 267     {
 268         std::copy(vowelA.begin(), vowelA.end(), std::begin(mChans[i].Formants[VOWEL_A_INDEX]));
 269         std::copy(vowelB.begin(), vowelB.end(), std::begin(mChans[i].Formants[VOWEL_B_INDEX]));
 270     }
 271
 272     mOutTarget = target.Main->Buffer;
 273     auto set_gains = [slot,target](auto &chan, al::span<const float,MaxAmbiChannels> coeffs)
 274     { ComputePanGains(target.Main, coeffs.data(), slot->Gain, chan.TargetGains); };
 275     SetAmbiPanIdentity(std::begin(mChans), slot->Wet.Buffer.size(), set_gains);
 276 }
 277
 278 void VmorpherState::process(const size_t samplesToDo, const al::span<const FloatBufferLine> samplesIn, const al::span<FloatBufferLine> samplesOut)
 279 {
 280     /* Following the EFX specification for a conformant implementation which describes
 281      * the effect as a pair of 4-band formant filters blended together using an LFO.
 282      */
 283     for(size_t base{0u};base < samplesToDo;)
 284     {
 285         const size_t td{minz(MAX_UPDATE_SAMPLES, samplesToDo-base)};
 286
 287         mGetSamples(mLfo, mIndex, mStep, td);
 288         mIndex += static_cast<uint>(mStep * td);
 289         mIndex &= WAVEFORM_FRACMASK;
 290
 291         auto chandata = std::begin(mChans);
 292         for(const auto &input : samplesIn)
 293         {
 294             auto& vowelA = chandata->Formants[VOWEL_A_INDEX];
 295             auto& vowelB = chandata->Formants[VOWEL_B_INDEX];
 296
 297             /* Process first vowel. */
 298             std::fill_n(std::begin(mSampleBufferA), td, 0.0f);
 299             vowelA[0].process(&input[base], mSampleBufferA, td);
 300             vowelA[1].process(&input[base], mSampleBufferA, td);
 301             vowelA[2].process(&input[base], mSampleBufferA, td);
 302             vowelA[3].process(&input[base], mSampleBufferA, td);
 303
 304             /* Process second vowel. */
 305             std::fill_n(std::begin(mSampleBufferB), td, 0.0f);
 306             vowelB[0].process(&input[base], mSampleBufferB, td);
 307             vowelB[1].process(&input[base], mSampleBufferB, td);
 308             vowelB[2].process(&input[base], mSampleBufferB, td);
 309             vowelB[3].process(&input[base], mSampleBufferB, td);
 310
 311             alignas(16) float blended[MAX_UPDATE_SAMPLES];
 312             for(size_t i{0u};i < td;i++)
 313                 blended[i] = lerp(mSampleBufferA[i], mSampleBufferB[i], mLfo[i]);
 314
 315             /* Now, mix the processed sound data to the output. */
 316             MixSamples({blended, td}, samplesOut, chandata->CurrentGains, chandata->TargetGains,
 317                 samplesToDo-base, base);
 318             ++chandata;
 319         }
 320
 321         base += td;
 322     }
 323 }
 324
 325
 326 struct VmorpherStateFactory final : public EffectStateFactory {
 327     al::intrusive_ptr<EffectState> create() override
 328     { return al::intrusive_ptr<EffectState>{new VmorpherState{}}; }
 329 };
 330
 331 } // namespace
 332
 333 EffectStateFactory *VmorpherStateFactory_getFactory()
 334 {
 335     static VmorpherStateFactory VmorpherFactory{};
 336     return &VmorpherFactory;
 337 }