vae::core::SpatialProcessor Class Reference

#include <vae_spatial_processor.hpp>

Collaboration diagram for vae::core::SpatialProcessor:

Public Member Functions
Result	init (Size hrtfVoices)
Size	mix (VoiceManger &manager, Bank &bank, SpatialManager &spatial, SampleIndex frames, Size sampleRate)
	Process a single bank. More...
Result	loadHRTF (const char *path, Size length, const char *rootPath, Size sampleRate)

Private Attributes
HRTF	mHRTF
	Currently loaded HRTF, there can only be one. More...
HRTFLoader	mHRTFLoader
	Struct to decode the hrtf. More...
HeapBuffer< VoiceHRTF >	mVoiceHRTFs
	Working data for convolution. More...
ScratchBuffer	mScratchBuffer
	Temporary filtered/looped signal TODO this will not work with parallel bank processing. More...

Detailed Description

Definition at line 20 of file vae_spatial_processor.hpp.

Member Function Documentation

init()

Result vae::core::SpatialProcessor::init ( Size  hrtfVoices )

inline

Definition at line 29 of file vae_spatial_processor.hpp.

                                     {
            VAE_PROFILER_SCOPE_NAMED("Spatial Processor Init")
            mVoiceHRTFs.resize(hrtfVoices);
            mScratchBuffer.resize(StaticConfig::MaxBlock);
            return Result::Success;
        }

Here is the call graph for this function:

loadHRTF()

Result vae::core::SpatialProcessor::loadHRTF ( const char *  path, Size  length, const char *  rootPath, Size  sampleRate  )

inline

Definition at line 282 of file vae_spatial_processor.hpp.

                                                                                              {
            Result result = mHRTFLoader.load(path, length, rootPath, sampleRate, mHRTF);
            if (result != Result::Success) { return result; }
            for (auto& i : mVoiceHRTFs) {
                i.convolutionBuffer.resize(mHRTF.irLength);
                i.convolutionBuffer.set();
            }
            return Result::Success;
        }

Here is the call graph for this function:

Here is the caller graph for this function:

mix()

Size vae::core::SpatialProcessor::mix ( VoiceManger &  manager, Bank &  bank, SpatialManager &  spatial, SampleIndex  frames, Size  sampleRate  )

inline

Process a single bank.

Parameters

manager
banks
frames
sampleRate

Returns

Number of voices mixed

Pan and mix templated lambda so we don't have to write this for each vonfig

Parameters

panner

Get's a panner instance with pan() and speakers() function

Definition at line 45 of file vae_spatial_processor.hpp.

          {
            Size actuallyRendered = 0;
            VAE_PROFILER_SCOPE_NAMED("Spatial Processor")
            manager.forEachVoice([&](Voice& v, Size vi) {
                if (v.bank != bank.id) { return true; }     // wrong bank
                if (!v.spatialized) { return true; }        // not spatialized
                VAE_PROFILER_SCOPE_NAMED("Spatial Voice")
                if (!spatial.hasEmitter(v.emitter)) {
                    VAE_DEBUG("Spatial voice is missing emitter")
                    return false; // ! needs emitter
                }
 
                auto& source = bank.sources[v.source];
                auto& signal = source.signal;
 
                const auto signalLength = signal.size();
 
                if (signalLength == 0) { return false; }    // ! no signal
 
                v.time = v.time % signalLength;     // Keep signal in bounds before starting
 
                if (signal.sampleRate != sampleRate) {
                    // VAE_DEBUG("Spatial Voice samplerate mismatch. Enabled filter.")
                    v.filtered = true; // implicitly filter to resample
                }
 
                auto& emitter = spatial.getEmitter(v.emitter);
                auto& target = bank.mixers[v.mixer].buffer;
                const Sample gain = v.gain * source.gain;
                auto& l = spatial.getListeners()[v.listener];
 
                Real distanceAttenuated;
                Vec3 relativeDirection;
                // * Attenuation calculation
                {
                    VAE_PROFILER_SCOPE_NAMED("Attenuation calculation")
                    // samething as graphics, make the world rotate round the listener
                    // TODO this should be possible without a 4x4 matrix?
                    glm::mat4x4 lookAt = glm::lookAt(l.position, l.position + l.front, l.up);
                    // listener is the world origin now
                    relativeDirection = (lookAt * glm::vec4(emitter.position, 1.f));
 
 
                    const Real distance = std::max(glm::length(relativeDirection), 0.1f);
                    relativeDirection /= distance;
 
                    if (v.attenuate) {
                        distanceAttenuated = distance;
                        distanceAttenuated = std::max(distanceAttenuated, Real(1)); // we don't want to get louder than 1
                        distanceAttenuated = Real(1) / distanceAttenuated;
                    } else {
                        distanceAttenuated = 1.0;
                    }
                    distanceAttenuated *= gain;
                }
 
                if (distanceAttenuated < StaticConfig::MinVolume) {
                    return true; // ! inaudible
                    // TODO maybe progress still progress time?
                }
                actuallyRendered++;
                target.setValidSize(frames); // mark mtarget ixer as active
                v.audible = true;
 
                // * Filtering and looping logic
 
                // TODO This thing is littered with branches, maybe needs some cleanup
 
                const Sample* in;               // The filtered, looped original signal used for panning later. We only do mono signals
                SampleIndex remaining = frames; // playback speed and looping affects this
                bool finished = false;          // the return value of this function stops the voice
 
                if (v.filtered) {
                    VAE_PROFILER_SCOPE_NAMED("Voice Filter")
                    auto& fd = manager.getVoiceFilter(vi);
 
                    if (!v.started) {
                        // Initialize filter variables when first playing the voice
                        fd.highpassScratch[0]   = 0;
                        fd.lowpassScratch[0]    = signal[0][v.time];
                    }
 
                    // Playback speed taking samplerate into account
                    const Sample speed = fd.speed * (Sample(signal.sampleRate) / Sample(sampleRate));
 
                    if (!v.loop) {
                        // If we're not looping, end time calculation is a bit more complex
                        remaining = std::min(
                            frames,
                            SampleIndex(std::floor((signalLength - v.time) / speed - fd.timeFract))
                        );
                        finished = remaining != frames; // we might have reached the end
                    }
 
                    // fractional time, we need the value after the loop, so it's defined outside
                    Real position;
                    for (SampleIndex s = 0; s < frames; s++) {
                        // Linear interpolation between two samples
                        position = v.time + (s * speed) + fd.timeFract;
                        const Real lastPosition = std::floor(position);
                        const Size lastIndex = (Size) lastPosition;
                        const Size nextIndex = (Size) lastPosition + 1;
 
                        Real mix = position - lastPosition;
                        // mix = 0.5 * (1.0 - cos((mix) * 3.1416)); // cosine interpolation, introduces new harmonics somehow
 
                        // TODO 30% of the time in here is spent on the modulo
                        const Sample last = signal[0][lastIndex % signalLength];
                        const Sample next = signal[0][nextIndex % signalLength];
                        // linear resampling, sounds alright enough
                        const Sample in = (last + mix * (next - last)) * gain;
 
                        //  * super simple lowpass and highpass filter
                        // just lerps with a previous value
                        const Sample lpd = in + fd.lowpass * (fd.lowpassScratch[0] - in);
                        fd.lowpassScratch[0] = lpd;
 
                        const Sample hps = fd.highpassScratch[0];
                        const Sample hpd = hps + fd.highpass * (in - hps);
                        fd.highpassScratch[0] = hpd;
 
                        mScratchBuffer[0][s] = (lpd - hpd);
                    }
                    position += speed;                  // step to next sample
                    v.time = (SampleIndex) std::floor(position);        // split the signal in normal sample position
                    fd.timeFract = position - v.time;   // and fractional time for the next block
                    v.time = v.time;                    // set index back
                    in = mScratchBuffer[0];             // set the buffer to use for panning
                } else {
                    VAE_PROFILER_SCOPE_NAMED("Non filtered Voice")
                    if (v.loop) {
                        // put the looped signal in scratch buffer eventhough we're not filtering
                        // so panning doesn't need to worry about looping
                        for (SampleIndex s = 0; s < frames; s++) {
                            mScratchBuffer[0][s] = signal[0][(v.time + s) % signalLength];
                        }
                        v.time = (v.time + frames); // progress the time
                        in = mScratchBuffer[0];     // set buffer for panning
                        finished = false;           // never stop the voice
                    } else {
                        // Not filtering or looping
                        // Means we can use the original signal buffer but need to
                        // set the remaining samples so we don't run over the signal end
                        remaining = std::min(
                            frames, SampleIndex(signalLength - v.time
                        ));
                        in = signal[0] + v.time;
                        finished = remaining != frames; // we might have reached the end
                        v.time += remaining;            // progress time in voice
                    }
                }
 
                if (l.configuration == Listener::Configuration::HRTF && v.HRTF && mHRTF.rate) {
                    // * HRTF Panning
                    VAE_ASSERT(vi < mVoiceHRTFs.size()) // only the lower voice can use hrtfs
                    VAE_PROFILER_SCOPE_NAMED("Render HRTF")
 
                    Size closestIndex = HRTFUtil::closest(mHRTF, relativeDirection);
 
                    if (closestIndex == ~Size(0)) { return true; } // ! no hrtf found?
 
                    auto& hrtfVoice = mVoiceHRTFs[vi];
 
                    if (!v.started) { // clear old data
                        hrtfVoice.convolutionIndex = 0;
                        hrtfVoice.convolutionBuffer.set();
                    }
 
                    HRTFUtil::apply(
                        mHRTF.positions[closestIndex],
                        hrtfVoice, remaining, target, in, distanceAttenuated
                    );
 
                } else {
                    VAE_PROFILER_SCOPE_NAMED("Render SPCAP")
                    // * Normal SPCAP panning
                    auto& lastPan = manager.getVoicePan(vi);
                    VoicePan currentPan;
                    auto& currentVolumes = currentPan.volumes;
                    auto& lastVolumes = lastPan.volumes;
 
                    /**
                    * @brief Pan and mix templated lambda so we don't have to write this for each vonfig
                    * @param panner Get's a panner instance with pan() and speakers() function
                    */
                    const auto pan = [&](const auto& panner) {
                        // This is actually constexpr but not according to clangd
                        constexpr Size channels = std::min(Size(StaticConfig::MaxChannels), panner.speakers);
                        panner.pan(
                            relativeDirection, currentVolumes,
                            distanceAttenuated, emitter.spread
                        );
 
                        if (!v.started) {
                            // first time don't interpolate
                            for (Size c = 0; c < channels; c++) {
                                lastVolumes[c] = currentVolumes[c];
                            }
                        }
 
                        Sample t = 0;
                        for (SampleIndex s = 0; s < remaining; s++) {
                            const Sample sample = in[s];
                            // lerp between last and current channel volumes
                            // Not correct in terms of power convservation, but easy and efficient
                            for (Size c = 0; c < channels; c++) {
                                target[c][s] += sample * (lastVolumes[c] + t * (currentVolumes[c] - lastVolumes[c]));
                            }
                            t += Sample(1) / Sample(frames);
                        }
                    };
 
                    switch (l.configuration) {
                        case Listener::Configuration::HRTF:
                        case Listener::Configuration::Headphones:   pan(SPCAP::HeadphoneSPCAP); break;
                        case Listener::Configuration::Stereo:       pan(SPCAP::StereroSPCAP);   break;
                        case Listener::Configuration::Suround:      pan(SPCAP::SuroundSPCAP);   break;
                        case Listener::Configuration::Quadrophonic: pan(SPCAP::QuadSPCAP);      break;
                        case Listener::Configuration::Mono:         pan(SPCAP::MonoSPCAP);      break;
                    }
 
                    lastPan = std::move(currentPan);
                }
                v.started = true;
                if (finished) {
                    emitter.autoplaying = false;
                    return false;
                }
                return true;
            });
            return actuallyRendered;
        }

Here is the call graph for this function:

Here is the caller graph for this function:

Member Data Documentation

mHRTF

HRTF vae::core::SpatialProcessor::mHRTF

private

Currently loaded HRTF, there can only be one.

Definition at line 21 of file vae_spatial_processor.hpp.

mHRTFLoader

HRTFLoader vae::core::SpatialProcessor::mHRTFLoader

private

Struct to decode the hrtf.

Definition at line 22 of file vae_spatial_processor.hpp.

mScratchBuffer

ScratchBuffer vae::core::SpatialProcessor::mScratchBuffer

private

Temporary filtered/looped signal TODO this will not work with parallel bank processing.

Definition at line 27 of file vae_spatial_processor.hpp.

mVoiceHRTFs

HeapBuffer<VoiceHRTF> vae::core::SpatialProcessor::mVoiceHRTFs

private

Working data for convolution.

Definition at line 23 of file vae_spatial_processor.hpp.

---

The documentation for this class was generated from the following file:

• vae_spatial_processor.hpp