#include "amuse/Voice.hpp" #include "amuse/Submix.hpp" #include "amuse/IBackendVoice.hpp" #include "amuse/IBackendVoiceAllocator.hpp" #include "amuse/AudioGroup.hpp" #include "amuse/Common.hpp" #include "amuse/Engine.hpp" #include "amuse/DSPCodec.h" #include "amuse/N64MusyXCodec.h" #include #include namespace amuse { extern "C" const float VolumeLUT[]; void Voice::_destroy() { Entity::_destroy(); for (std::shared_ptr& vox : m_childVoices) vox->_destroy(); } Voice::~Voice() { // fprintf(stderr, "DEALLOC %d\n", m_vid); } Voice::Voice(Engine& engine, const AudioGroup& group, int groupId, int vid, bool emitter, std::weak_ptr studio) : Entity(engine, group, groupId), m_vid(vid), m_emitter(emitter), m_studio(studio) { // fprintf(stderr, "ALLOC %d\n", m_vid); } Voice::Voice(Engine& engine, const AudioGroup& group, int groupId, ObjectId oid, int vid, bool emitter, std::weak_ptr studio) : Entity(engine, group, groupId, oid), m_vid(vid), m_emitter(emitter), m_studio(studio) { // fprintf(stderr, "ALLOC %d\n", m_vid); } void Voice::_macroSampleEnd() { if (m_sampleEndTrap.macroId != 0xffff) { if (m_sampleEndTrap.macroId == m_state.m_header.m_macroId) { m_state.m_pc.back().second = m_sampleEndTrap.macroStep; m_state.m_inWait = false; } else loadSoundObject(m_sampleEndTrap.macroId, m_sampleEndTrap.macroStep, m_state.m_ticksPerSec, m_state.m_initKey, m_state.m_initVel, m_state.m_initMod); } else m_state.sampleEndNotify(*this); } bool Voice::_checkSamplePos(bool& looped) { looped = false; if (!m_curSample) return true; if (m_curSamplePos >= m_lastSamplePos) { if (m_curSample->first.m_loopLengthSamples) { /* Turn over looped sample */ m_curSamplePos = m_curSample->first.m_loopStartSample; if (m_curFormat == SampleFormat::DSP) { m_prev1 = m_curSample->second.dsp.m_hist1; m_prev2 = m_curSample->second.dsp.m_hist2; } looped = true; } else { /* Notify sample end */ _macroSampleEnd(); m_curSample = nullptr; return true; } } /* Looped samples issue sample end when ADSR envelope complete */ if (m_volAdsr.isComplete()) { _macroSampleEnd(); m_curSample = nullptr; return true; } return false; } void Voice::_doKeyOff() { if (m_state.m_inWait && m_state.m_keyoffWait) { if (m_volAdsr.isAdsrSet() || m_state.m_useAdsrControllers) m_volAdsr.keyOff(*this); if (m_pitchAdsr.isAdsrSet()) m_pitchAdsr.keyOff(); } else { m_volAdsr.keyOff(*this); m_pitchAdsr.keyOff(); } m_state.keyoffNotify(*this); } void Voice::_setTotalPitch(int32_t cents, bool slew) { // fprintf(stderr, "PITCH %d %d \n", cents, slew); int32_t interval = cents - m_curSample->first.m_pitch * 100; double ratio = std::exp2(interval / 1200.0); m_sampleRate = m_curSample->first.m_sampleRate * ratio; m_backendVoice->setPitchRatio(ratio, slew); } bool Voice::_isRecursivelyDead() { if (m_voxState != VoiceState::Dead) return false; for (std::shared_ptr& vox : m_childVoices) if (!vox->_isRecursivelyDead()) return false; return true; } void Voice::_bringOutYourDead() { for (auto it = m_childVoices.begin(); it != m_childVoices.end();) { Voice* vox = it->get(); vox->_bringOutYourDead(); if (vox->_isRecursivelyDead()) { it = _destroyVoice(it); continue; } ++it; } } std::shared_ptr Voice::_findVoice(int vid, std::weak_ptr thisPtr) { if (m_vid == vid) return thisPtr.lock(); for (std::shared_ptr& vox : m_childVoices) { std::shared_ptr ret = vox->_findVoice(vid, vox); if (ret) return ret; } return {}; } std::unique_ptr& Voice::_ensureCtrlVals() { if (m_ctrlValsSelf) return m_ctrlValsSelf; m_ctrlValsSelf.reset(new int8_t[128]); memset(m_ctrlValsSelf.get(), 0, 128); return m_ctrlValsSelf; } std::list>::iterator Voice::_allocateVoice(double sampleRate, bool dynamicPitch) { auto it = m_childVoices.emplace( m_childVoices.end(), new Voice(m_engine, m_audioGroup, m_groupId, m_engine.m_nextVid++, m_emitter, m_studio)); m_childVoices.back()->m_backendVoice = m_engine.getBackend().allocateVoice(*m_childVoices.back(), sampleRate, dynamicPitch); return it; } std::list>::iterator Voice::_destroyVoice(std::list>::iterator it) { if ((*it)->m_destroyed) return m_childVoices.begin(); (*it)->_destroy(); return m_childVoices.erase(it); } template static T ApplyVolume(float vol, T samp) { /* -10dB to 0dB mapped to full volume range */ return samp * VolumeLUT[int(vol * 65536)]; } void Voice::_procSamplePre(int16_t& samp) { double dt; /* Block linearized will use a larger `dt` for amplitude sampling; * significantly reducing the processing expense */ switch (m_engine.m_ampMode) { case AmplitudeMode::PerSample: m_voiceSamples += 1; dt = 1.0 / m_sampleRate; break; case AmplitudeMode::BlockLinearized: { uint32_t rem = m_voiceSamples % 160; m_voiceSamples += 1; dt = m_sampleRate * 160; if (rem != 0) { /* Lerp within 160-sample block */ float t = rem / 160.f; float l = clamp(0.f, m_lastLevel * (1.f - t) + m_nextLevel * t, 1.f); /* Apply total volume to sample using decibel scale */ samp = ApplyVolume(l, samp); return; } dt = 160.0 / m_sampleRate; break; } } m_voiceTime += dt; /* Process active envelope */ if (m_envelopeTime >= 0.0) { m_envelopeTime += dt; float start = m_envelopeStart; float end = m_envelopeEnd; float t = clamp(0.f, float(m_envelopeTime / m_envelopeDur), 1.f); if (m_envelopeCurve) t = (*m_envelopeCurve)[int(t * 127.f)] / 127.f; m_curVol = clamp(0.f, (start * (1.0f - t)) + (end * t), 1.f); // printf("%d %f\n", m_vid, m_curVol); /* Done with envelope */ if (m_envelopeTime > m_envelopeDur) m_envelopeTime = -1.f; } /* Dynamically evaluate per-sample SoundMacro parameters */ /* Process user volume slew */ if (m_engine.m_ampMode == AmplitudeMode::PerSample) { if (m_targetUserVol != m_curUserVol) { float samplesPer5Ms = m_sampleRate * 5.f / 1000.f; if (samplesPer5Ms > 1.f) { float adjRate = 1.f / samplesPer5Ms; if (m_targetUserVol < m_curUserVol) { m_curUserVol -= adjRate; if (m_targetUserVol > m_curUserVol) m_curUserVol = m_targetUserVol; } else { m_curUserVol += adjRate; if (m_targetUserVol < m_curUserVol) m_curUserVol = m_targetUserVol; } } else m_curUserVol = m_targetUserVol; } } else m_curUserVol = m_targetUserVol; /* Factor in ADSR envelope state */ float adsr = m_volAdsr.advance(dt, *this); m_lastLevel = m_nextLevel; m_nextLevel = m_curUserVol * m_curVol * adsr * (m_state.m_curVel / 127.f); /* Apply tremolo */ if (m_state.m_tremoloSel && (m_tremoloScale || m_tremoloModScale)) { float t = m_state.m_tremoloSel.evaluate(m_voiceTime, *this, m_state) / 2.f; if (m_tremoloScale && m_tremoloModScale) { float fac = (1.0f - t) + (m_tremoloScale * t); float modT = m_state.m_modWheelSel ? (m_state.m_modWheelSel.evaluate(m_voiceTime, *this, m_state) / 2.f) : (getCtrlValue(1) / 127.f); float modFac = (1.0f - modT) + (m_tremoloModScale * modT); m_nextLevel *= fac * modFac; } else if (m_tremoloScale) { float fac = (1.0f - t) + (m_tremoloScale * t); m_nextLevel *= fac; } else if (m_tremoloModScale) { float modT = m_state.m_modWheelSel ? (m_state.m_modWheelSel.evaluate(m_voiceTime, *this, m_state) / 2.f) : (getCtrlValue(1) / 127.f); float modFac = (1.0f - modT) + (m_tremoloModScale * modT); m_nextLevel *= modFac; } } m_nextLevel = clamp(0.f, m_nextLevel, 1.f); /* Apply total volume to sample using decibel scale */ samp = ApplyVolume(m_nextLevel, samp); } template T Voice::_procSampleMaster(double time, T samp) { float evalVol = m_state.m_volumeSel ? (m_state.m_volumeSel.evaluate(time, *this, m_state) / 2.f) : 1.f; return ApplyVolume(clamp(0.f, evalVol, 1.f), samp); } template T Voice::_procSampleAuxA(double time, T samp) { float evalVol = m_state.m_volumeSel ? (m_state.m_volumeSel.evaluate(time, *this, m_state) / 2.f) : 1.f; evalVol *= m_state.m_reverbSel ? (m_state.m_reverbSel.evaluate(time, *this, m_state) / 2.f) : m_curReverbVol; evalVol += m_state.m_preAuxASel ? (m_state.m_preAuxASel.evaluate(time, *this, m_state) / 2.f) : 0.f; return ApplyVolume(clamp(0.f, evalVol, 1.f), samp); } template T Voice::_procSampleAuxB(double time, T samp) { float evalVol = m_state.m_volumeSel ? (m_state.m_volumeSel.evaluate(time, *this, m_state) / 2.f) : 1.f; evalVol *= m_state.m_postAuxB ? (m_state.m_postAuxB.evaluate(time, *this, m_state) / 2.f) : m_curAuxBVol; evalVol += m_state.m_preAuxBSel ? (m_state.m_preAuxBSel.evaluate(time, *this, m_state) / 2.f) : 0.f; return ApplyVolume(clamp(0.f, evalVol, 1.f), samp); } uint32_t Voice::_GetBlockSampleCount(SampleFormat fmt) { switch (fmt) { default: return 1; case Voice::SampleFormat::DSP: return 14; case Voice::SampleFormat::N64: return 64; } } void Voice::preSupplyAudio(double dt) { /* Process SoundMacro; bootstrapping sample if needed */ bool dead = m_state.advance(*this, dt); /* Process per-block evaluators here */ if (m_state.m_pedalSel) { bool pedal = m_state.m_pedalSel.evaluate(m_voiceTime, *this, m_state) >= 1.f; if (pedal != m_sustained) setPedal(pedal); } bool panDirty = false; if (m_state.m_panSel) { float evalPan = m_state.m_panSel.evaluate(m_voiceTime, *this, m_state); if (evalPan != m_curPan) { m_curPan = evalPan; panDirty = true; } } if (m_state.m_spanSel) { float evalSpan = m_state.m_spanSel.evaluate(m_voiceTime, *this, m_state); if (evalSpan != m_curSpan) { m_curSpan = evalSpan; panDirty = true; } } if (panDirty) _setPan(m_curPan); if (m_state.m_pitchWheelSel) _setPitchWheel(m_state.m_pitchWheelSel.evaluate(m_voiceTime, *this, m_state)); /* Process active pan-sweep */ bool refresh = false; if (m_panningTime >= 0.f) { m_panningTime += dt; float start = (m_panPos - 64) / 64.f; float end = (m_panPos + m_panWidth - 64) / 64.f; float t = clamp(0.f, m_panningTime / m_panningDur, 1.f); _setPan((start * (1.0f - t)) + (end * t)); refresh = true; /* Done with panning */ if (m_panningTime > m_panningDur) m_panningTime = -1.f; } /* Process active span-sweep */ if (m_spanningTime >= 0.f) { m_spanningTime += dt; float start = (m_spanPos - 64) / 64.f; float end = (m_spanPos + m_spanWidth - 64) / 64.f; float t = clamp(0.f, m_spanningTime / m_spanningDur, 1.f); _setSurroundPan((start * (1.0f - t)) + (end * t)); refresh = true; /* Done with spanning */ if (m_spanningTime > m_spanningDur) m_spanningTime = -1.f; } /* Calculate total pitch */ int32_t newPitch = m_curPitch; refresh |= m_pitchDirty; m_pitchDirty = false; if (m_portamentoTime >= 0.f) { m_portamentoTime += dt; float t = clamp(0.f, m_portamentoTime / m_state.m_portamentoTime, 1.f); newPitch = (m_curPitch * (1.0f - t)) + (m_portamentoTarget * t); refresh = true; /* Done with portamento */ if (m_portamentoTime > m_state.m_portamentoTime) { m_portamentoTime = -1.f; m_curPitch = m_portamentoTarget; } } if (m_pitchEnv) { newPitch *= m_pitchAdsr.advance(dt); refresh = true; } /* Process pitch sweep 1 */ if (m_pitchSweep1It < m_pitchSweep1Times) { ++m_pitchSweep1It; m_pitchSweep1 = m_pitchSweep1Add * m_pitchSweep1It / m_pitchSweep1Times; refresh = true; } /* Process pitch sweep 2 */ if (m_pitchSweep2It < m_pitchSweep2Times) { ++m_pitchSweep2It; m_pitchSweep2 = m_pitchSweep2Add * m_pitchSweep2It / m_pitchSweep2Times; refresh = true; } if (m_curSample && refresh) { _setTotalPitch(newPitch + m_pitchSweep1 + m_pitchSweep2 + m_pitchWheelVal, m_needsSlew); m_needsSlew = true; } if (dead && (!m_curSample || m_voxState == VoiceState::KeyOff) && m_sampleEndTrap.macroId == 0xffff && m_messageTrap.macroId == 0xffff && (!m_curSample || (m_curSample && m_volAdsr.isComplete()))) { m_voxState = VoiceState::Dead; m_backendVoice->stop(); } } size_t Voice::supplyAudio(size_t samples, int16_t* data) { uint32_t samplesRem = samples; if (m_curSample) { uint32_t blockSampleCount = _GetBlockSampleCount(m_curFormat); uint32_t block; bool looped = true; while (looped && samplesRem) { block = m_curSamplePos / blockSampleCount; uint32_t rem = m_curSamplePos % blockSampleCount; if (rem) { uint32_t remCount = std::min(samplesRem, m_lastSamplePos - block * blockSampleCount); uint32_t decSamples; switch (m_curFormat) { case SampleFormat::DSP: { decSamples = DSPDecompressFrameRanged(data, m_curSampleData + 8 * block, m_curSample->second.dsp.m_coefs, &m_prev1, &m_prev2, rem, remCount); break; } case SampleFormat::N64: { decSamples = N64MusyXDecompressFrameRanged(data, m_curSampleData + 256 + 40 * block, m_curSample->second.vadpcm.m_coefs, rem, remCount); break; } case SampleFormat::PCM: { const int16_t* pcm = reinterpret_cast(m_curSampleData); remCount = std::min(samplesRem, m_lastSamplePos - m_curSamplePos); for (uint32_t i = 0; i < remCount; ++i) data[i] = SBig(pcm[m_curSamplePos + i]); decSamples = remCount; break; } case SampleFormat::PCM_PC: { const int16_t* pcm = reinterpret_cast(m_curSampleData); remCount = std::min(samplesRem, m_lastSamplePos - m_curSamplePos); memmove(data, pcm + m_curSamplePos, remCount * sizeof(int16_t)); decSamples = remCount; break; } default: memset(data, 0, sizeof(int16_t) * samples); return samples; } /* Per-sample processing */ for (uint32_t i = 0; i < decSamples; ++i) { ++m_curSamplePos; _procSamplePre(data[i]); } samplesRem -= decSamples; data += decSamples; } if (_checkSamplePos(looped)) { if (samplesRem) memset(data, 0, sizeof(int16_t) * samplesRem); return samples; } if (looped) continue; while (samplesRem) { block = m_curSamplePos / blockSampleCount; uint32_t remCount = std::min(samplesRem, m_lastSamplePos - block * blockSampleCount); uint32_t decSamples; switch (m_curFormat) { case SampleFormat::DSP: { decSamples = DSPDecompressFrame(data, m_curSampleData + 8 * block, m_curSample->second.dsp.m_coefs, &m_prev1, &m_prev2, remCount); break; } case SampleFormat::N64: { decSamples = N64MusyXDecompressFrame(data, m_curSampleData + 256 + 40 * block, m_curSample->second.vadpcm.m_coefs, remCount); break; } case SampleFormat::PCM: { const int16_t* pcm = reinterpret_cast(m_curSampleData); remCount = std::min(samplesRem, m_lastSamplePos - m_curSamplePos); for (uint32_t i = 0; i < remCount; ++i) data[i] = SBig(pcm[m_curSamplePos + i]); decSamples = remCount; break; } case SampleFormat::PCM_PC: { const int16_t* pcm = reinterpret_cast(m_curSampleData); remCount = std::min(samplesRem, m_lastSamplePos - m_curSamplePos); memmove(data, pcm + m_curSamplePos, remCount * sizeof(int16_t)); decSamples = remCount; break; } default: memset(data, 0, sizeof(int16_t) * samples); return samples; } /* Per-sample processing */ for (uint32_t i = 0; i < decSamples; ++i) { ++m_curSamplePos; _procSamplePre(data[i]); } samplesRem -= decSamples; data += decSamples; if (_checkSamplePos(looped)) { if (samplesRem) memset(data, 0, sizeof(int16_t) * samplesRem); return samples; } if (looped) break; } } } else memset(data, 0, sizeof(int16_t) * samples); if (m_voxState == VoiceState::Dead) m_curSample = nullptr; return samples; } void Voice::routeAudio(size_t frames, double dt, int busId, int16_t* in, int16_t* out) { dt /= double(frames); switch (busId) { case 0: default: { for (uint32_t i = 0; i < frames; ++i) out[i] = _procSampleMaster(dt * i + m_voiceTime, in[i]); break; } case 1: { for (uint32_t i = 0; i < frames; ++i) out[i] = _procSampleAuxA(dt * i + m_voiceTime, in[i]); break; } case 2: { for (uint32_t i = 0; i < frames; ++i) out[i] = _procSampleAuxB(dt * i + m_voiceTime, in[i]); break; } } } void Voice::routeAudio(size_t frames, double dt, int busId, int32_t* in, int32_t* out) { dt /= double(frames); switch (busId) { case 0: default: { for (uint32_t i = 0; i < frames; ++i) out[i] = _procSampleMaster(dt * i + m_voiceTime, in[i]); break; } case 1: { for (uint32_t i = 0; i < frames; ++i) out[i] = _procSampleAuxA(dt * i + m_voiceTime, in[i]); break; } case 2: { for (uint32_t i = 0; i < frames; ++i) out[i] = _procSampleAuxB(dt * i + m_voiceTime, in[i]); break; } } } void Voice::routeAudio(size_t frames, double dt, int busId, float* in, float* out) { dt /= double(frames); switch (busId) { case 0: default: { for (uint32_t i = 0; i < frames; ++i) out[i] = _procSampleMaster(dt * i + m_voiceTime, in[i]); break; } case 1: { for (uint32_t i = 0; i < frames; ++i) out[i] = _procSampleAuxA(dt * i + m_voiceTime, in[i]); break; } case 2: { for (uint32_t i = 0; i < frames; ++i) out[i] = _procSampleAuxB(dt * i + m_voiceTime, in[i]); break; } } } int Voice::maxVid() const { int maxVid = m_vid; for (const std::shared_ptr& vox : m_childVoices) maxVid = std::max(maxVid, vox->maxVid()); return maxVid; } std::shared_ptr Voice::_startChildMacro(ObjectId macroId, int macroStep, double ticksPerSec, uint8_t midiKey, uint8_t midiVel, uint8_t midiMod, bool pushPc) { std::list>::iterator vox = _allocateVoice(32000.0, true); if (!(*vox)->loadSoundObject(macroId, macroStep, ticksPerSec, midiKey, midiVel, midiMod, pushPc)) { _destroyVoice(vox); return {}; } (*vox)->setVolume(m_targetUserVol); (*vox)->setPan(m_userPan); (*vox)->setSurroundPan(m_userSpan); return *vox; } std::shared_ptr Voice::startChildMacro(int8_t addNote, ObjectId macroId, int macroStep) { return _startChildMacro(macroId, macroStep, 1000.0, m_state.m_initKey + addNote, m_state.m_initVel, m_state.m_initMod); } bool Voice::_loadSoundMacro(const unsigned char* macroData, int macroStep, double ticksPerSec, uint8_t midiKey, uint8_t midiVel, uint8_t midiMod, bool pushPc) { if (m_state.m_pc.empty()) m_state.initialize(macroData, macroStep, ticksPerSec, midiKey, midiVel, midiMod, m_audioGroup.getDataFormat() != DataFormat::PC); else { if (!pushPc) m_state.m_pc.clear(); m_state.m_pc.push_back({macroData, macroStep}); m_state.m_header = *reinterpret_cast(macroData); if (m_audioGroup.getDataFormat() != DataFormat::PC) m_state.m_header.swapBig(); } m_voxState = VoiceState::Playing; m_backendVoice->start(); return true; } bool Voice::_loadKeymap(const Keymap* keymap, int macroStep, double ticksPerSec, uint8_t midiKey, uint8_t midiVel, uint8_t midiMod, bool pushPc) { const Keymap& km = keymap[midiKey]; ObjectId oid = (m_audioGroup.getDataFormat() == DataFormat::PC) ? km.objectId : SBig(km.objectId); midiKey += km.transpose; bool ret = loadSoundObject(oid, macroStep, ticksPerSec, midiKey, midiVel, midiMod, pushPc); m_curVol = 1.f; _setPan((km.pan - 64) / 64.f); _setSurroundPan(-1.f); return ret; } bool Voice::_loadLayer(const std::vector& layer, int macroStep, double ticksPerSec, uint8_t midiKey, uint8_t midiVel, uint8_t midiMod, bool pushPc) { bool ret = false; for (const LayerMapping* mapping : layer) { if (midiKey >= mapping->keyLo && midiKey <= mapping->keyHi) { ObjectId oid = (m_audioGroup.getDataFormat() == DataFormat::PC) ? mapping->objectId : SBig(mapping->objectId); uint8_t mappingKey = midiKey + mapping->transpose; if (m_voxState != VoiceState::Playing) { ret |= loadSoundObject(oid, macroStep, ticksPerSec, mappingKey, midiVel, midiMod, pushPc); m_curVol = mapping->volume / 127.f; _setPan((mapping->pan - 64) / 64.f); _setSurroundPan((mapping->span - 64) / 64.f); } else { std::shared_ptr vox = _startChildMacro(oid, macroStep, ticksPerSec, mappingKey, midiVel, midiMod, pushPc); if (vox) { vox->m_curVol = mapping->volume / 127.f; vox->_setPan((mapping->pan - 64) / 64.f); vox->_setSurroundPan((mapping->span - 64) / 64.f); ret = true; } } } } return ret; } bool Voice::loadSoundObject(ObjectId objectId, int macroStep, double ticksPerSec, uint8_t midiKey, uint8_t midiVel, uint8_t midiMod, bool pushPc) { const unsigned char* macroData = m_audioGroup.getPool().soundMacro(objectId); if (macroData) return _loadSoundMacro(macroData, macroStep, ticksPerSec, midiKey, midiVel, midiMod, pushPc); const Keymap* keymap = m_audioGroup.getPool().keymap(objectId); if (keymap) return _loadKeymap(keymap, macroStep, ticksPerSec, midiKey, midiVel, midiMod, pushPc); const std::vector* layer = m_audioGroup.getPool().layer(objectId); if (layer) return _loadLayer(*layer, macroStep, ticksPerSec, midiKey, midiVel, midiMod, pushPc); return false; } void Voice::_macroKeyOff() { if (m_voxState == VoiceState::Playing) { if (m_sustained) m_sustainKeyOff = true; else _doKeyOff(); m_voxState = VoiceState::KeyOff; } } void Voice::keyOff() { if (m_keyoffTrap.macroId != 0xffff) { if (m_keyoffTrap.macroId == m_state.m_header.m_macroId) { m_state.m_pc.back().second = m_keyoffTrap.macroStep; m_state.m_inWait = false; } else loadSoundObject(m_keyoffTrap.macroId, m_keyoffTrap.macroStep, m_state.m_ticksPerSec, m_state.m_initKey, m_state.m_initVel, m_state.m_initMod); } else _macroKeyOff(); for (const std::shared_ptr& vox : m_childVoices) vox->keyOff(); } void Voice::message(int32_t val) { m_messageQueue.push_back(val); if (m_messageTrap.macroId != 0xffff) { if (m_messageTrap.macroId == m_state.m_header.m_macroId) m_state.m_pc.back().second = m_messageTrap.macroStep; else loadSoundObject(m_messageTrap.macroId, m_messageTrap.macroStep, m_state.m_ticksPerSec, m_state.m_initKey, m_state.m_initVel, m_state.m_initMod); } } void Voice::startSample(int16_t sampId, int32_t offset) { m_curSample = m_audioGroup.getSample(sampId); if (m_curSample) { m_sampleRate = m_curSample->first.m_sampleRate; m_curPitch = m_curSample->first.m_pitch; m_pitchDirty = true; _setPitchWheel(m_curPitchWheel); m_backendVoice->resetSampleRate(m_curSample->first.m_sampleRate); m_needsSlew = false; int32_t numSamples = m_curSample->first.m_numSamples & 0xffffff; if (offset) { if (m_curSample->first.m_loopLengthSamples) { if (offset > int32_t(m_curSample->first.m_loopStartSample)) offset = ((offset - m_curSample->first.m_loopStartSample) % m_curSample->first.m_loopLengthSamples) + m_curSample->first.m_loopStartSample; } else offset = clamp(0, offset, numSamples); } m_curSamplePos = offset; m_curSampleData = m_audioGroup.getSampleData(m_curSample->first.m_sampleOff); m_prev1 = 0; m_prev2 = 0; if (m_audioGroup.getDataFormat() == DataFormat::PC) m_curFormat = SampleFormat::PCM_PC; else m_curFormat = SampleFormat(m_curSample->first.m_numSamples >> 24); if (m_curFormat == SampleFormat::DSP_DRUM) m_curFormat = SampleFormat::DSP; m_lastSamplePos = m_curSample->first.m_loopLengthSamples ? (m_curSample->first.m_loopStartSample + m_curSample->first.m_loopLengthSamples) : numSamples; bool looped; _checkSamplePos(looped); /* Seek DSPADPCM state if needed */ if (m_curSample && m_curSamplePos && m_curFormat == SampleFormat::DSP) { uint32_t block = m_curSamplePos / 14; uint32_t rem = m_curSamplePos % 14; for (uint32_t b = 0; b < block; ++b) DSPDecompressFrameStateOnly(m_curSampleData + 8 * b, m_curSample->second.dsp.m_coefs, &m_prev1, &m_prev2, 14); if (rem) DSPDecompressFrameStateOnly(m_curSampleData + 8 * block, m_curSample->second.dsp.m_coefs, &m_prev1, &m_prev2, rem); } } } void Voice::stopSample() { m_curSample = nullptr; } void Voice::setVolume(float vol) { m_targetUserVol = clamp(0.f, vol, 1.f); for (std::shared_ptr& vox : m_childVoices) vox->setVolume(vol); } void Voice::_setPan(float pan) { m_curPan = clamp(-1.f, pan, 1.f); float totalPan = clamp(-1.f, m_curPan + m_userPan, 1.f); float totalSpan = clamp(-1.f, m_curSpan + m_userSpan, 1.f); float coefs[8]; /* Left */ coefs[0] = (totalPan <= 0.f) ? 1.f : (1.f - totalPan); coefs[0] *= (totalSpan <= 0.f) ? 1.f : (1.f - totalSpan); /* Right */ coefs[1] = (totalPan >= 0.f) ? 1.f : (1.f + totalPan); coefs[1] *= (totalSpan <= 0.f) ? 1.f : (1.f - totalSpan); /* Back Left */ coefs[2] = (totalPan <= 0.f) ? 1.f : (1.f - totalPan); coefs[2] *= (totalSpan >= 0.f) ? 1.f : (1.f + totalSpan); /* Back Right */ coefs[3] = (totalPan >= 0.f) ? 1.f : (1.f + totalPan); coefs[3] *= (totalSpan >= 0.f) ? 1.f : (1.f + totalSpan); /* Center */ coefs[4] = 1.f - std::fabs(totalPan); /* LFE */ coefs[5] = 1.f; /* Side Left */ coefs[6] = (totalPan <= 0.f) ? 1.f : (1.f - totalPan); coefs[6] *= 1.f - std::fabs(totalSpan); /* Side Right */ coefs[7] = (totalPan >= 0.f) ? 1.f : (1.f + totalPan); coefs[7] *= 1.f - std::fabs(totalSpan); m_backendVoice->setChannelLevels(m_studio->getMaster().m_backendSubmix.get(), coefs, true); m_backendVoice->setChannelLevels(m_studio->getAuxA().m_backendSubmix.get(), coefs, true); m_backendVoice->setChannelLevels(m_studio->getAuxB().m_backendSubmix.get(), coefs, true); } void Voice::setPan(float pan) { m_userPan = pan; _setPan(m_curPan); for (std::shared_ptr& vox : m_childVoices) vox->setPan(pan); } void Voice::_setSurroundPan(float span) { m_curSpan = clamp(-1.f, span, 1.f); _setPan(m_curPan); } void Voice::setSurroundPan(float span) { m_userSpan = span; _setSurroundPan(m_curSpan); for (std::shared_ptr& vox : m_childVoices) vox->setSurroundPan(span); } void Voice::startEnvelope(double dur, float vol, const Curve* envCurve) { m_envelopeTime = 0.f; m_envelopeDur = dur; m_envelopeStart = clamp(0.f, m_curVol, 1.f); m_envelopeEnd = clamp(0.f, vol, 1.f); m_envelopeCurve = envCurve; } void Voice::startFadeIn(double dur, float vol, const Curve* envCurve) { m_envelopeTime = 0.f; m_envelopeDur = dur; m_envelopeStart = 0.f; m_envelopeEnd = clamp(0.f, vol, 1.f); m_envelopeCurve = envCurve; } void Voice::startPanning(double dur, uint8_t panPos, int8_t panWidth) { m_panningTime = 0.f; m_panningDur = dur; m_panPos = panPos; m_panWidth = panWidth; } void Voice::startSpanning(double dur, uint8_t spanPos, int8_t spanWidth) { m_spanningTime = 0.f; m_spanningDur = dur; m_spanPos = spanPos; m_spanWidth = spanWidth; } void Voice::setPitchKey(int32_t cents) { m_curPitch = cents; m_pitchDirty = true; } void Voice::setPedal(bool pedal) { if (m_sustained && !pedal && m_sustainKeyOff) { m_sustainKeyOff = false; _doKeyOff(); } m_sustained = pedal; for (std::shared_ptr& vox : m_childVoices) vox->setPedal(pedal); } void Voice::setDoppler(float) {} void Voice::setVibrato(int32_t level, int32_t modLevel, float period) { m_vibratoLevel = level; m_vibratoModLevel = modLevel; m_vibratoPeriod = period; } void Voice::setMod2VibratoRange(int32_t modLevel) { m_vibratoModLevel = modLevel; } void Voice::setTremolo(float tremoloScale, float tremoloModScale) { m_tremoloScale = tremoloScale; m_tremoloModScale = tremoloModScale; } void Voice::setPitchSweep1(uint8_t times, int16_t add) { m_pitchSweep1 = 0; m_pitchSweep1It = 0; m_pitchSweep1Times = times; m_pitchSweep1Add = add; } void Voice::setPitchSweep2(uint8_t times, int16_t add) { m_pitchSweep2 = 0; m_pitchSweep2It = 0; m_pitchSweep2Times = times; m_pitchSweep2Add = add; } void Voice::setReverbVol(float rvol) { m_curReverbVol = clamp(0.f, rvol, 1.f); for (std::shared_ptr& vox : m_childVoices) vox->setReverbVol(rvol); } void Voice::setAuxBVol(float bvol) { m_curAuxBVol = clamp(0.f, bvol, 1.f); for (std::shared_ptr& vox : m_childVoices) vox->setAuxBVol(bvol); } void Voice::setAdsr(ObjectId adsrId, bool dls) { if (dls) { const ADSRDLS* adsr = m_audioGroup.getPool().tableAsAdsrDLS(adsrId); if (adsr) { m_volAdsr.reset(adsr, m_state.m_initKey, m_state.m_initVel); if (m_voxState == VoiceState::KeyOff) m_volAdsr.keyOff(*this); } } else { const ADSR* adsr = m_audioGroup.getPool().tableAsAdsr(adsrId); if (adsr) { m_volAdsr.reset(adsr); if (m_voxState == VoiceState::KeyOff) m_volAdsr.keyOff(*this); } } } void Voice::setPitchFrequency(uint32_t hz, uint16_t fine) { m_sampleRate = hz + fine / 65536.0; m_backendVoice->setPitchRatio(1.0, false); m_backendVoice->resetSampleRate(m_sampleRate); } void Voice::setPitchAdsr(ObjectId adsrId, int32_t cents) { const ADSRDLS* adsr = m_audioGroup.getPool().tableAsAdsrDLS(adsrId); if (adsr) { m_pitchAdsr.reset(adsr, m_state.m_initKey, m_state.m_initVel); m_pitchEnvRange = cents; m_pitchEnv = true; } } void Voice::_setPitchWheel(float pitchWheel) { if (pitchWheel > 0.f) m_pitchWheelVal = m_pitchWheelUp * m_curPitchWheel; else if (pitchWheel < 0.f) m_pitchWheelVal = m_pitchWheelDown * m_curPitchWheel; else m_pitchWheelVal = 0; m_pitchDirty = true; } void Voice::setPitchWheel(float pitchWheel) { m_curPitchWheel = amuse::clamp(-1.f, pitchWheel, 1.f); _setPitchWheel(m_curPitchWheel); for (std::shared_ptr& vox : m_childVoices) vox->setPitchWheel(pitchWheel); } void Voice::setPitchWheelRange(int8_t up, int8_t down) { m_pitchWheelUp = up * 100; m_pitchWheelDown = down * 100; _setPitchWheel(m_curPitchWheel); } void Voice::setAftertouch(uint8_t aftertouch) { m_curAftertouch = aftertouch; for (std::shared_ptr& vox : m_childVoices) vox->setAftertouch(aftertouch); } bool Voice::doPortamento(uint8_t newNote) { bool pState; switch (m_state.m_portamentoMode) { case 0: default: pState = false; break; case 1: pState = true; break; case 2: pState = m_state.m_portamentoSel ? (m_state.m_portamentoSel.evaluate(m_voiceTime, *this, m_state) >= 1.f) : (getCtrlValue(65) >= 64); break; } if (!pState) return false; m_portamentoTime = 0.f; m_portamentoTarget = newNote * 100; m_state.m_initKey = newNote; return true; } void Voice::_notifyCtrlChange(uint8_t ctrl, int8_t val) { if (ctrl == 0x40) { if (val >= 0x40) setPedal(true); else setPedal(false); } else if (ctrl == 0x5b) { setReverbVol(val / 127.f); } else if (ctrl == 0x5d) { setAuxBVol(val / 127.f); } for (std::shared_ptr& vox : m_childVoices) vox->_notifyCtrlChange(ctrl, val); } size_t Voice::getTotalVoices() const { size_t ret = 1; for (const std::shared_ptr& vox : m_childVoices) ret += vox->getTotalVoices(); return ret; } void Voice::kill() { m_voxState = VoiceState::Dead; m_backendVoice->stop(); for (const std::shared_ptr& vox : m_childVoices) vox->kill(); } }