Okay, starting to looks seriosly good, almost works on all platforms too now! lol

src/AudioEngine.cpp

@@ -279,59 +279,77 @@ void AudioEngine::onFinished() {
   // Notify UI that track is ready to play
   emit trackLoaded(true);
 
-  // Emit immediately so analyzer can use pcmData fallback while Hilbert runs
+  // Emit early with pcmData-only so analyzer can show spectrum immediately.
+  // The Hilbert task builds a NEW TrackData, so no data race.
   emit trackDataChanged(m_trackData);
 
-  // OPTIMIZATION: Run heavy analysis in background to avoid blocking audio
+  // Run heavy analysis in background thread pool
-  // thread FIX: Use QPointer to prevent crash if AudioEngine is deleted
-  // before task runs
   QPointer<AudioEngine> self = this;
-  QThreadPool::globalInstance()->start([self, newData]() {
+  // Capture pcmData via implicit sharing (cheap refcount bump)
+  QByteArray pcmSnap = newData->pcmData;
+  int sr = newData->sampleRate;
+  QThreadPool::globalInstance()->start([self, pcmSnap, sr]() {
     if (!self)
       return;
 
     const float *rawFloats =
-        reinterpret_cast<const float *>(newData->pcmData.constData());
+        reinterpret_cast<const float *>(pcmSnap.constData());
-    long long totalFloats = newData->pcmData.size() / sizeof(float);
+    long long totalFloats = pcmSnap.size() / sizeof(float);
     long long totalFrames = totalFloats / 2;
 
-    if (totalFrames > 0) {
+    if (totalFrames <= 0)
-      // 1. BPM Detection
+      return;
+
+    // 1. BPM Detection
 #ifdef ENABLE_TEMPO_ESTIMATION
-      MemoryAudioReader reader(rawFloats, totalFrames, newData->sampleRate);
+    MemoryAudioReader reader(rawFloats, totalFrames, sr);
-      auto bpmOpt =
+    auto bpmOpt =
-          LTE::GetBpm(reader, LTE::FalsePositiveTolerance::Lenient, nullptr);
+        LTE::GetBpm(reader, LTE::FalsePositiveTolerance::Lenient, nullptr);
-
+    float bpm = bpmOpt.has_value() ? static_cast<float>(*bpmOpt) : 0.0f;
-      // Emit BPM result back to main thread context
+    if (self) {
-      float bpm = bpmOpt.has_value() ? static_cast<float>(*bpmOpt) : 0.0f;
+      QMetaObject::invokeMethod(self, "analysisReady", Qt::QueuedConnection,
-
+                                Q_ARG(float, bpm), Q_ARG(float, 1.0f));
-      if (self) {
+    }
-        QMetaObject::invokeMethod(self, "analysisReady", Qt::QueuedConnection,
-                                  Q_ARG(float, bpm), Q_ARG(float, 1.0f));
-      }
 #endif
 
-      // 2. Hilbert Transform
+    // 2. Hilbert Transform — process one channel at a time to minimize
-      std::vector<double> inputL(totalFrames), inputR(totalFrames);
+    // peak memory. FFTW uses 4 buffers per channel instead of 8.
-      for (size_t i = 0; i < totalFrames; ++i) {
+    auto finalData = std::make_shared<TrackData>();
-        inputL[i] = static_cast<double>(rawFloats[i * 2]);
+    finalData->pcmData = pcmSnap;
-        inputR[i] = static_cast<double>(rawFloats[i * 2 + 1]);
+    finalData->sampleRate = sr;
-      }
+    finalData->valid = true;
-      BlockHilbert blockHilbert;
+    finalData->complexData.resize(totalFloats);
-      auto analyticPair = blockHilbert.hilbertTransform(inputL, inputR);
 
-      newData->complexData.resize(totalFloats);
+    BlockHilbert blockHilbert;
-      for (size_t i = 0; i < totalFrames; ++i) {
+    // Reusable input buffer (one channel at a time)
-        newData->complexData[i * 2] = analyticPair.first[i];
+    std::vector<double> input(totalFrames);
-        newData->complexData[i * 2 + 1] = analyticPair.second[i];
-      }
 
-      // Notify Analyzer that complex data is ready
+    // Left channel: build input, transform, copy to complexData, free result
-      if (self) {
+    for (size_t i = 0; i < static_cast<size_t>(totalFrames); ++i)
-        QMetaObject::invokeMethod(self, "trackDataChanged",
+      input[i] = static_cast<double>(rawFloats[i * 2]);
-                                  Qt::QueuedConnection,
+    {
-                                  Q_ARG(std::shared_ptr<TrackData>, newData));
+      auto analytic = blockHilbert.hilbertTransformSingle(input);
-      }
+      for (size_t i = 0; i < static_cast<size_t>(totalFrames); ++i)
+        finalData->complexData[i * 2] = analytic[i];
+    } // analytic freed
+
+    // Right channel: reuse input buffer
+    for (size_t i = 0; i < static_cast<size_t>(totalFrames); ++i)
+      input[i] = static_cast<double>(rawFloats[i * 2 + 1]);
+    {
+      auto analytic = blockHilbert.hilbertTransformSingle(input);
+      for (size_t i = 0; i < static_cast<size_t>(totalFrames); ++i)
+        finalData->complexData[i * 2 + 1] = analytic[i];
+    } // analytic freed
+
+    // Free input buffer
+    { std::vector<double>().swap(input); }
+
+    // Notify Analyzer with the complete data
+    if (self) {
+      QMetaObject::invokeMethod(self, "trackDataChanged",
+                                Qt::QueuedConnection,
+                                Q_ARG(std::shared_ptr<TrackData>, finalData));
     }
   });
 }
@@ -591,7 +609,11 @@ void AudioAnalyzer::processLoop() {
         specMain.db[b] = val;
       }
     }
-    results.push_back({specMain.freqs, specMain.db, primaryDb});
+    FrameData fd{specMain.freqs, specMain.db, primaryDb, {}};
+    // Pass cepstrum from ch0 main processor only (mono is sufficient)
+    if (i == 0)
+      fd.cepstrum = std::move(specMain.cepstrum);
+    results.push_back(std::move(fd));
   }
 
   // 6. Publish Result

src/AudioEngine.h

@@ -211,6 +211,7 @@ public:
     std::vector<float> freqs;
     std::vector<float> db;
     std::vector<float> primaryDb;
+    std::vector<float> cepstrum;  // from main processor ch0 only
   };
 
   // Thread-safe pull for UI

src/MainWindow.cpp

@@ -92,8 +92,8 @@ MainWindow::MainWindow(QWidget *parent) : QMainWindow(parent) {
 
   // Settings -> Analyzer
   connect(m_playerPage->settings(), &SettingsWidget::paramsChanged, this,
-          [this](bool, bool, bool, bool, float, float, float, int granularity,
+          [this](bool, bool, bool, bool, bool, float, float, float,
-                 int detail, float strength) {
+                 int granularity, int detail, float strength) {
             QMetaObject::invokeMethod(
                 m_analyzer, "setSmoothingParams", Qt::QueuedConnection,
                 Q_ARG(int, granularity), Q_ARG(int, detail),
@@ -411,10 +411,10 @@ void MainWindow::loadSettings() {
     m_playerPage->settings()->setParams(
         root["glass"].toBool(true), root["focus"].toBool(false),
         root["albumColors"].toBool(false), root["mirrored"].toBool(false),
-        root["bins"].toInt(26), root["fps"].toInt(60),
+        root["inverted"].toBool(false), root["bins"].toInt(26),
-        root["brightness"].toDouble(1.0), root["granularity"].toInt(33),
+        root["fps"].toInt(60), root["brightness"].toDouble(1.0),
-        root["detail"].toInt(50), root["strength"].toDouble(0.0),
+        root["granularity"].toInt(33), root["detail"].toInt(50),
-        root["bpmScaleIndex"].toInt(2));
+        root["strength"].toDouble(0.0), root["bpmScaleIndex"].toInt(2));
   }
 }
 
@@ -427,6 +427,7 @@ void MainWindow::saveSettings() {
   root["focus"] = s->isFocus();
   root["albumColors"] = s->isAlbumColors();
   root["mirrored"] = s->isMirrored();
+  root["inverted"] = s->isInverted();
   root["bins"] = s->getBins();
   root["fps"] = s->getFps();
   root["brightness"] = s->getBrightness();
@@ -489,7 +490,7 @@ void MainWindow::updateSmoothing() {
   float targetStrength = 0.8f * (1.0f - normalized);
   SettingsWidget *s = m_playerPage->settings();
   s->setParams(s->isGlass(), s->isFocus(), s->isAlbumColors(), s->isMirrored(),
-               s->getBins(), s->getFps(), s->getBrightness(),
+               s->isInverted(), s->getBins(), s->getFps(), s->getBrightness(),
                s->getGranularity(), s->getDetail(), targetStrength,
                s->getBpmScaleIndex());
 }

src/PlayerControls.cpp

@@ -136,6 +136,7 @@ SettingsWidget::SettingsWidget(QWidget *parent) : QWidget(parent) {
   m_checkFocus = createCheck("Focus", true, 0, 1);
   m_checkAlbumColors = createCheck("Album Colors", false, 1, 0);
   m_checkMirrored = createCheck("Mirrored", true, 1, 1);
+  m_checkInverted = createCheck("Invert", false, 2, 0);
   layout->addLayout(grid);
 
   // Helper for sliders
@@ -237,7 +238,7 @@ SettingsWidget::SettingsWidget(QWidget *parent) : QWidget(parent) {
 }
 
 void SettingsWidget::setParams(bool glass, bool focus, bool albumColors,
-                               bool mirrored, int bins, int fps,
+                               bool mirrored, bool inverted, int bins, int fps,
                                float brightness, int granularity, int detail,
                                float strength, int bpmScaleIndex) {
   bool oldState = blockSignals(true);
@@ -245,6 +246,7 @@ void SettingsWidget::setParams(bool glass, bool focus, bool albumColors,
   m_checkFocus->setChecked(focus);
   m_checkAlbumColors->setChecked(albumColors);
   m_checkMirrored->setChecked(mirrored);
+  m_checkInverted->setChecked(inverted);
   m_sliderBins->setValue(bins);
   m_lblBins->setText(QString("Bins: %1").arg(bins));
 
@@ -278,7 +280,8 @@ void SettingsWidget::setParams(bool glass, bool focus, bool albumColors,
 void SettingsWidget::emitParams() {
   emit paramsChanged(m_checkGlass->isChecked(), m_checkFocus->isChecked(),
                      m_checkAlbumColors->isChecked(),
-                     m_checkMirrored->isChecked(), m_hue, m_contrast,
+                     m_checkMirrored->isChecked(),
+                     m_checkInverted->isChecked(), m_hue, m_contrast,
                      m_brightness, m_granularity, m_detail, m_strength);
 }
 

src/PlayerControls.h

@@ -45,6 +45,7 @@ public:
   bool isFocus() const { return m_checkFocus->isChecked(); }
   bool isAlbumColors() const { return m_checkAlbumColors->isChecked(); }
   bool isMirrored() const { return m_checkMirrored->isChecked(); }
+  bool isInverted() const { return m_checkInverted->isChecked(); }
   int getBins() const { return m_sliderBins->value(); }
   int getFps() const { return m_sliderFps->value(); }
   float getBrightness() const { return m_brightness; }
@@ -58,13 +59,15 @@ public:
   int getBpmScaleIndex() const;
 
   void setParams(bool glass, bool focus, bool albumColors, bool mirrored,
-                 int bins, int fps, float brightness, int granularity,
+                 bool inverted, int bins, int fps, float brightness,
-                 int detail, float strength, int bpmScaleIndex);
+                 int granularity, int detail, float strength,
+                 int bpmScaleIndex);
 
 signals:
   void paramsChanged(bool glass, bool focus, bool albumColors, bool mirrored,
-                     float hue, float contrast, float brightness,
+                     bool inverted, float hue, float contrast,
-                     int granularity, int detail, float strength);
+                     float brightness, int granularity, int detail,
+                     float strength);
   void fpsChanged(int fps);
   void dspParamsChanged(int fft, int hop);
   void binsChanged(int n);
@@ -85,6 +88,7 @@ private:
   QCheckBox *m_checkFocus;
   QCheckBox *m_checkAlbumColors;
   QCheckBox *m_checkMirrored;
+  QCheckBox *m_checkInverted;
   XYPad *m_padDsp;
   XYPad *m_padColor;
   QSlider *m_sliderBins;

src/Processor.cpp

@@ -271,30 +271,38 @@ Processor::Spectrum Processor::getSpectrum() {
         freqsFull[i] = freq;
     }
 
+    // --- Cepstral IFFT (always compute for cepstrum visualization) ---
+    // 1. Log Magnitude
+    for(int i=0; i<m_frameSize; ++i) {
+        // Mirror for symmetry to get real cepstrum
+        int idx = (i <= m_frameSize/2) ? i : (m_frameSize - i);
+        double val = std::max(1e-9, magFull[idx]);
+        m_cep_in[i][0] = std::log(val);
+        m_cep_in[i][1] = 0.0;
+    }
+
+    // 2. IFFT -> Cepstrum
+    fftw_execute(m_cep_plan_inv); // Result in m_cep_out (scaled by N)
+
+    // 3. Extract positive quefrencies for visualization
+    int halfN = m_frameSize / 2;
+    std::vector<float> cepCoeffs(halfN);
+    double cepScale = 1.0 / m_frameSize;
+    for(int i=0; i<halfN; ++i) {
+        cepCoeffs[i] = static_cast<float>(m_cep_out[i][0] * cepScale);
+    }
+
     // --- Cepstral Smoothing / Trig Interpolation ---
     if (m_cepstralStrength > 0.0f) {
-        // 1. Log Magnitude
+        // Hilbert on Cepstrum (Analytic Cepstrum)
-        for(int i=0; i<m_frameSize; ++i) {
+        m_cep_in[0][0] = m_cep_out[0][0] * cepScale;
-            // Mirror for symmetry to get real cepstrum
+        m_cep_in[0][1] = m_cep_out[0][1] * cepScale;
-            int idx = (i <= m_frameSize/2) ? i : (m_frameSize - i);
-            double val = std::max(1e-9, magFull[idx]);
-            m_cep_in[i][0] = std::log(val);
-            m_cep_in[i][1] = 0.0;
-        }
 
-        // 2. IFFT -> Cepstrum
-        fftw_execute(m_cep_plan_inv); // Result in m_cep_out (scaled by N)
-        
-        // 3. Hilbert on Cepstrum (Analytic Cepstrum)
-        double scale = 1.0 / m_frameSize;
-        m_cep_in[0][0] = m_cep_out[0][0] * scale; 
-        m_cep_in[0][1] = m_cep_out[0][1] * scale;
-        
         for(int i=1; i<m_frameSize; ++i) {
             if (i < m_frameSize/2) {
                 // Positive quefrencies * 2
-                m_cep_in[i][0] = m_cep_out[i][0] * scale * 2.0;
+                m_cep_in[i][0] = m_cep_out[i][0] * cepScale * 2.0;
-                m_cep_in[i][1] = m_cep_out[i][1] * scale * 2.0;
+                m_cep_in[i][1] = m_cep_out[i][1] * cepScale * 2.0;
             } else {
                 // Negative quefrencies = 0
                 m_cep_in[i][0] = 0.0;
@@ -302,9 +310,9 @@ Processor::Spectrum Processor::getSpectrum() {
             }
         }
 
-        // 4. Idealize Curve (Smoothing)
+        // Idealize Curve (Smoothing)
         std::vector<double> envelope = idealizeCurve(magFull);
-        
+
         // Apply Strength (Mix)
         for(size_t i=0; i<magFull.size(); ++i) {
             magFull[i] = magFull[i] * (1.0 - m_cepstralStrength) + envelope[i] * m_cepstralStrength;
@@ -351,5 +359,5 @@ Processor::Spectrum Processor::getSpectrum() {
     std::vector<float> freqsRet(m_freqsConst.size());
     for(size_t i=0; i<m_freqsConst.size(); ++i) freqsRet[i] = static_cast<float>(m_freqsConst[i]);
 
-    return {freqsRet, averagedDb};
+    return {freqsRet, averagedDb, cepCoeffs};
 }

src/Processor.h

@@ -29,6 +29,7 @@ public:
     struct Spectrum {
         std::vector<float> freqs;
         std::vector<float> db;
+        std::vector<float> cepstrum;  // raw cepstral coefficients (positive quefrencies)
     };
 
     Spectrum getSpectrum();

src/VisualizerWidget.cpp

@@ -45,12 +45,13 @@ void VisualizerWidget::setTargetFps(int fps) {
 }
 
 void VisualizerWidget::setParams(bool glass, bool focus, bool albumColors,
-                                 bool mirrored, float hue, float contrast,
+                                 bool mirrored, bool inverted, float hue,
-                                 float brightness) {
+                                 float contrast, float brightness) {
   m_glass = glass;
   m_focus = focus;
   m_useAlbumColors = albumColors;
   m_mirrored = mirrored;
+  m_inverted = inverted;
   m_hueFactor = hue;
   m_contrast = contrast;
   m_brightness = brightness;
@@ -287,6 +288,16 @@ void VisualizerWidget::updateData(
       b.cachedColor = binColor;
     }
   }
+
+  // --- 5. Cepstral Thread Smoothing (mirrored mode only) ---
+  if (m_mirrored && !data.empty() && !data[0].cepstrum.empty()) {
+    const auto &raw = data[0].cepstrum;
+    if (m_smoothedCepstrum.size() != raw.size())
+      m_smoothedCepstrum.assign(raw.size(), 0.0f);
+    for (size_t i = 0; i < raw.size(); ++i)
+      m_smoothedCepstrum[i] = 0.15f * raw[i] + 0.85f * m_smoothedCepstrum[i];
+  }
+
   update();
 }
 
@@ -307,10 +318,10 @@ void VisualizerWidget::initialize(QRhiCommandBuffer *cb) {
                                   2048 * 6 * sizeof(float)));
     m_vbuf->create();
 
-    // Uniform buffer: 4 aligned MVP matrices (for mirrored mode)
+    // Uniform buffer: 5 aligned MVP matrices (4 mirror passes + 1 cepstrum)
     m_ubuf.reset(m_rhi->newBuffer(QRhiBuffer::Dynamic,
                                   QRhiBuffer::UniformBuffer,
-                                  m_ubufAlign * 4));
+                                  m_ubufAlign * 5));
     m_ubuf->create();
 
     // Shader resource bindings with dynamic UBO offset
@@ -376,10 +387,13 @@ void VisualizerWidget::render(QRhiCommandBuffer *cb) {
   // Only rebuild vertices when new data has arrived
   if (m_dataDirty) {
     m_dataDirty = false;
-    if (m_mirrored)
+    if (m_mirrored) {
-      buildVertices(w / 2, h / 2);
+      buildVertices(w * 0.55f, h / 2);
-    else
+      buildCepstrumVertices(w, h);
+    } else {
       buildVertices(w, h);
+      m_cepstrumVertexCount = 0;
+    }
   }
 
   int numPasses = m_mirrored ? 4 : 1;
@@ -387,14 +401,21 @@ void VisualizerWidget::render(QRhiCommandBuffer *cb) {
   // Prepare resource updates
   QRhiResourceUpdateBatch *u = m_rhi->nextResourceUpdateBatch();
 
-  // Upload vertex data
+  // Upload vertex data (main + cepstrum appended)
-  if (!m_vertices.empty()) {
+  {
-    int dataSize = static_cast<int>(m_vertices.size() * sizeof(float));
+    int mainSize = static_cast<int>(m_vertices.size() * sizeof(float));
-    if (dataSize > m_vbuf->size()) {
+    int cepSize = static_cast<int>(m_cepstrumVerts.size() * sizeof(float));
-      m_vbuf->setSize(dataSize);
+    int totalSize = mainSize + cepSize;
-      m_vbuf->create();
+    if (totalSize > 0) {
+      if (totalSize > m_vbuf->size()) {
+        m_vbuf->setSize(totalSize);
+        m_vbuf->create();
+      }
+      if (mainSize > 0)
+        u->updateDynamicBuffer(m_vbuf.get(), 0, mainSize, m_vertices.data());
+      if (cepSize > 0)
+        u->updateDynamicBuffer(m_vbuf.get(), mainSize, cepSize, m_cepstrumVerts.data());
     }
-    u->updateDynamicBuffer(m_vbuf.get(), 0, dataSize, m_vertices.data());
   }
 
   // Upload MVP matrices
@@ -427,6 +448,15 @@ void VisualizerWidget::render(QRhiCommandBuffer *cb) {
                            mvp.constData());
   }
 
+  // Upload full-screen ortho MVP for cepstrum (slot 4)
+  if (m_mirrored && m_cepstrumVertexCount > 0) {
+    QMatrix4x4 cepProj;
+    cepProj.ortho(0, (float)w, (float)h, 0, -1, 1);
+    QMatrix4x4 cepMvp = correction * cepProj;
+    u->updateDynamicBuffer(m_ubuf.get(), 4 * m_ubufAlign, 64,
+                           cepMvp.constData());
+  }
+
   // Begin render pass
   cb->beginPass(renderTarget(), QColor(0, 0, 0, 255), {1.0f, 0}, u);
   cb->setViewport({0, 0, (float)outputSize.width(),
@@ -452,6 +482,15 @@ void VisualizerWidget::render(QRhiCommandBuffer *cb) {
     }
   }
 
+  // --- Cepstral Thread (single full-screen pass, after mirror loop) ---
+  if (m_mirrored && m_cepstrumVertexCount > 0) {
+    QRhiCommandBuffer::DynamicOffset cepOfs(0, quint32(4 * m_ubufAlign));
+    cb->setGraphicsPipeline(m_linePipeline.get());
+    cb->setShaderResources(m_srb.get(), 1, &cepOfs);
+    cb->setVertexInput(0, 1, &vbufBinding);
+    cb->draw(m_cepstrumVertexCount, 1, m_fillVertexCount + m_lineVertexCount, 0);
+  }
+
   cb->endPass();
   update();
 }
@@ -464,6 +503,76 @@ void VisualizerWidget::releaseResources() {
   m_vbuf.reset();
 }
 
+// ===== Cepstral Thread Vertex Building =====
+
+void VisualizerWidget::buildCepstrumVertices(int w, int h) {
+  m_cepstrumVerts.clear();
+  m_cepstrumVertexCount = 0;
+
+  if (m_smoothedCepstrum.empty())
+    return;
+
+  // Quefrency range: indices 12-600 (~80Hz to ~4000Hz pitch at 48kHz)
+  int qStart = 12;
+  int qEnd = std::min(600, (int)m_smoothedCepstrum.size());
+  if (qEnd <= qStart)
+    return;
+
+  // Find peak magnitude for normalization
+  float peak = 0.0f;
+  for (int i = qStart; i < qEnd; ++i)
+    peak = std::max(peak, std::abs(m_smoothedCepstrum[i]));
+  if (peak < 1e-7f)
+    return; // silence — don't draw
+
+  float invPeak = 1.0f / peak;
+  float maxDisp = w * 0.06f;
+  float cx = w * 0.5f;
+
+  // Color: unified color desaturated slightly, alpha ~0.45
+  float cr, cg, cb;
+  {
+    float ch, cs, cv;
+    m_unifiedColor.getHsvF(&ch, &cs, &cv);
+    cs *= 0.7f; // desaturate
+    QColor c = QColor::fromHsvF(ch, cs, cv);
+    cr = c.redF();
+    cg = c.greenF();
+    cb = c.blueF();
+  }
+  float ca = 0.45f;
+
+  // Build line segments with top/bottom edge fade
+  float fadeMargin = 0.08f; // fade over 8% of height at each end
+  float prevX = cx + m_smoothedCepstrum[qStart] * invPeak * maxDisp;
+  float prevY = 0.0f;
+  float prevT = 0.0f;
+
+  for (int i = qStart + 1; i < qEnd; ++i) {
+    float t = (float)(i - qStart) / (qEnd - qStart);
+    float y = t * h;
+    float x = cx + m_smoothedCepstrum[i] * invPeak * maxDisp;
+
+    // Fade alpha near top and bottom edges
+    auto edgeFade = [&](float tt) -> float {
+      if (tt < fadeMargin) return tt / fadeMargin;
+      if (tt > 1.0f - fadeMargin) return (1.0f - tt) / fadeMargin;
+      return 1.0f;
+    };
+    float a0 = ca * edgeFade(prevT);
+    float a1 = ca * edgeFade(t);
+
+    m_cepstrumVerts.insert(m_cepstrumVerts.end(),
+                           {prevX, prevY, cr, cg, cb, a0,
+                            x, y, cr, cg, cb, a1});
+    prevX = x;
+    prevY = y;
+    prevT = t;
+  }
+
+  m_cepstrumVertexCount = (int)m_cepstrumVerts.size() / 6;
+}
+
 // ===== Vertex Building (identical logic to old drawContent) =====
 
 void VisualizerWidget::buildVertices(int w, int h) {
@@ -483,11 +592,13 @@ void VisualizerWidget::buildVertices(int w, int h) {
 
     float xOffset = (ch == 1 && m_data.size() > 1) ? 1.005f : 1.0f;
 
-    for (size_t i = 0; i + 1 < freqs.size(); ++i) {
+    size_t numBins = std::min(freqs.size(), bins.size());
-      if (i + 1 >= bins.size())
+    for (size_t i = 0; i + 1 < numBins; ++i) {
-        break;
+      // When inverted, read bin data in reverse order (highs left, lows right)
-      const auto &b = bins[i];
+      size_t di = m_inverted ? (numBins - 1 - i) : i;
-      const auto &bNext = bins[i + 1];
+      size_t diN = m_inverted ? (numBins - 2 - i) : (i + 1);
+      const auto &b = bins[di];
+      const auto &bNext = bins[diN];
 
       // --- Brightness ---
       float avgEnergy =
@@ -527,8 +638,19 @@ void VisualizerWidget::buildVertices(int w, int h) {
       float alpha = 0.4f + (avgEnergy - 0.5f) * m_contrast;
       alpha = std::clamp(alpha * aMult, 0.0f, 1.0f);
 
+      // --- Edge fade: taper last bins to transparent near center gap ---
+      if (m_mirrored) {
+        int fadeBins = 4;
+        int fromEnd = (int)(numBins - 2) - (int)i;
+        if (fromEnd < fadeBins) {
+          float fade = (float)(fromEnd + 1) / (float)(fadeBins + 1);
+          fade = fade * fade; // ease-in for smoother taper
+          alpha *= fade;
+        }
+      }
+
       fillColor.setAlphaF(alpha);
-      lineColor.setAlphaF(0.9f);
+      lineColor.setAlphaF(std::min(0.9f, alpha));
 
       // --- Channel 1 tint ---
       if (ch == 1 && m_data.size() > 1) {

src/VisualizerWidget.h

@@ -15,7 +15,7 @@ class VisualizerWidget : public QRhiWidget {
 public:
     VisualizerWidget(QWidget* parent = nullptr);
     void updateData(const std::vector<AudioAnalyzer::FrameData>& data);
-    void setParams(bool glass, bool focus, bool albumColors, bool mirrored, float hue, float contrast, float brightness);
+    void setParams(bool glass, bool focus, bool albumColors, bool mirrored, bool inverted, float hue, float contrast, float brightness);
     void setAlbumPalette(const std::vector<QColor>& palette);
     void setNumBins(int n);
     void setTargetFps(int fps);
@@ -60,6 +60,7 @@ private:
     bool m_focus = false;
     bool m_useAlbumColors = false;
     bool m_mirrored = false;
+    bool m_inverted = false;
     float m_hueFactor = 0.9f;
     float m_contrast = 1.0f;
     float m_brightness = 1.0f;
@@ -82,6 +83,12 @@ private:
     int m_fillVertexCount = 0;
     int m_lineVertexCount = 0;
 
+    // Cepstral thread visualization
+    std::vector<float> m_smoothedCepstrum;
+    std::vector<float> m_cepstrumVerts;
+    int m_cepstrumVertexCount = 0;
+    void buildCepstrumVertices(int w, int h);
+
     float getX(float freq);
     QColor applyModifiers(QColor c);
 };

src/complex_block.cpp

@@ -32,71 +32,69 @@
 #include <stdexcept>
 #include <cmath>
 
-std::pair<std::vector<std::complex<double>>, std::vector<std::complex<double>>>
+std::vector<std::complex<double>>
-BlockHilbert::hilbertTransform(const std::vector<double>& left_signal, const std::vector<double>& right_signal) {
+BlockHilbert::hilbertTransformSingle(const std::vector<double>& signal) {
-    size_t n = left_signal.size();
+    size_t n = signal.size();
-    if (n == 0 || n != right_signal.size()) {
+    if (n == 0) return {};
-        return {{}, {}};
-    }
 
-    fftw_complex* fft_in_L = (fftw_complex*)fftw_malloc(sizeof(fftw_complex) * n);
+    fftw_complex* fft_in  = (fftw_complex*)fftw_malloc(sizeof(fftw_complex) * n);
-    fftw_complex* fft_out_L = (fftw_complex*)fftw_malloc(sizeof(fftw_complex) * n);
+    fftw_complex* fft_out = (fftw_complex*)fftw_malloc(sizeof(fftw_complex) * n);
-    fftw_complex* ifft_in_L = (fftw_complex*)fftw_malloc(sizeof(fftw_complex) * n);
+    fftw_complex* ifft_in = (fftw_complex*)fftw_malloc(sizeof(fftw_complex) * n);
-    fftw_complex* ifft_out_L = (fftw_complex*)fftw_malloc(sizeof(fftw_complex) * n);
+    fftw_complex* ifft_out= (fftw_complex*)fftw_malloc(sizeof(fftw_complex) * n);
 
-    fftw_complex* fft_in_R = (fftw_complex*)fftw_malloc(sizeof(fftw_complex) * n);
+    if (!fft_in || !fft_out || !ifft_in || !ifft_out) {
-    fftw_complex* fft_out_R = (fftw_complex*)fftw_malloc(sizeof(fftw_complex) * n);
+        fftw_free(fft_in); fftw_free(fft_out); fftw_free(ifft_in); fftw_free(ifft_out);
-    fftw_complex* ifft_in_R = (fftw_complex*)fftw_malloc(sizeof(fftw_complex) * n);
-    fftw_complex* ifft_out_R = (fftw_complex*)fftw_malloc(sizeof(fftw_complex) * n);
-
-    if (!fft_in_L || !fft_out_L || !ifft_in_L || !ifft_out_L || !fft_in_R || !fft_out_R || !ifft_in_R || !ifft_out_R) {
         throw std::runtime_error("FFTW memory allocation failed in BlockHilbert.");
     }
 
-    fftw_plan plan_forward_L = fftw_plan_dft_1d(static_cast<int>(n), fft_in_L, fft_out_L, FFTW_FORWARD, FFTW_ESTIMATE);
+    fftw_plan plan_fwd = fftw_plan_dft_1d(static_cast<int>(n), fft_in, fft_out, FFTW_FORWARD, FFTW_ESTIMATE);
-    fftw_plan plan_backward_L = fftw_plan_dft_1d(static_cast<int>(n), ifft_in_L, ifft_out_L, FFTW_BACKWARD, FFTW_ESTIMATE);
+    fftw_plan plan_bwd = fftw_plan_dft_1d(static_cast<int>(n), ifft_in, ifft_out, FFTW_BACKWARD, FFTW_ESTIMATE);
-    fftw_plan plan_forward_R = fftw_plan_dft_1d(static_cast<int>(n), fft_in_R, fft_out_R, FFTW_FORWARD, FFTW_ESTIMATE);
-    fftw_plan plan_backward_R = fftw_plan_dft_1d(static_cast<int>(n), ifft_in_R, ifft_out_R, FFTW_BACKWARD, FFTW_ESTIMATE);
 
-    if (!plan_forward_L || !plan_backward_L || !plan_forward_R || !plan_backward_R) {
+    if (!plan_fwd || !plan_bwd) {
-        fftw_free(fft_in_L); fftw_free(fft_out_L); fftw_free(ifft_in_L); fftw_free(ifft_out_L);
+        fftw_free(fft_in); fftw_free(fft_out); fftw_free(ifft_in); fftw_free(ifft_out);
-        fftw_free(fft_in_R); fftw_free(fft_out_R); fftw_free(ifft_in_R); fftw_free(ifft_out_R);
         throw std::runtime_error("FFTW plan creation failed in BlockHilbert.");
     }
 
     for (size_t i = 0; i < n; ++i) {
-        fft_in_L[i][0] = left_signal[i]; fft_in_L[i][1] = 0.0;
+        fft_in[i][0] = signal[i]; fft_in[i][1] = 0.0;
-        fft_in_R[i][0] = right_signal[i]; fft_in_R[i][1] = 0.0;
     }
 
-    fftw_execute(plan_forward_L);
+    fftw_execute(plan_fwd);
-    fftw_execute(plan_forward_R);
 
     for (size_t i = 0; i < n; ++i) {
         double multiplier = 1.0;
-        if (i > 0 && i < n / 2.0) { multiplier = 2.0; } 
+        if (i > 0 && i < n / 2.0) { multiplier = 2.0; }
         else if (i > n / 2.0) { multiplier = 0.0; }
-        
+        ifft_in[i][0] = fft_out[i][0] * multiplier;
-        ifft_in_L[i][0] = fft_out_L[i][0] * multiplier; ifft_in_L[i][1] = fft_out_L[i][1] * multiplier;
+        ifft_in[i][1] = fft_out[i][1] * multiplier;
-        ifft_in_R[i][0] = fft_out_R[i][0] * multiplier; ifft_in_R[i][1] = fft_out_R[i][1] * multiplier;
     }
 
-    fftw_execute(plan_backward_L);
+    fftw_execute(plan_bwd);
-    fftw_execute(plan_backward_R);
 
-    std::vector<std::complex<double>> analytic_L(n);
+    std::vector<std::complex<double>> result(n);
-    std::vector<std::complex<double>> analytic_R(n);
+    double inv_n = 1.0 / static_cast<double>(n);
     for (size_t i = 0; i < n; ++i) {
-        analytic_L[i].real(ifft_out_L[i][0] / static_cast<double>(n));
+        result[i].real(ifft_out[i][0] * inv_n);
-        analytic_L[i].imag(ifft_out_L[i][1] / static_cast<double>(n));
+        result[i].imag(ifft_out[i][1] * inv_n);
-        analytic_R[i].real(ifft_out_R[i][0] / static_cast<double>(n));
-        analytic_R[i].imag(ifft_out_R[i][1] / static_cast<double>(n));
     }
 
-    fftw_destroy_plan(plan_forward_L); fftw_destroy_plan(plan_backward_L);
+    fftw_destroy_plan(plan_fwd);
-    fftw_destroy_plan(plan_forward_R); fftw_destroy_plan(plan_backward_R);
+    fftw_destroy_plan(plan_bwd);
-    fftw_free(fft_in_L); fftw_free(fft_out_L); fftw_free(ifft_in_L); fftw_free(ifft_out_L);
+    fftw_free(fft_in); fftw_free(fft_out); fftw_free(ifft_in); fftw_free(ifft_out);
-    fftw_free(fft_in_R); fftw_free(fft_out_R); fftw_free(ifft_in_R); fftw_free(ifft_out_R);
 
-    return {analytic_L, analytic_R};
+    return result;
+}
+
+std::pair<std::vector<std::complex<double>>, std::vector<std::complex<double>>>
+BlockHilbert::hilbertTransform(const std::vector<double>& left_signal, const std::vector<double>& right_signal) {
+    if (left_signal.empty() || left_signal.size() != right_signal.size()) {
+        return {{}, {}};
+    }
+
+    // Process channels sequentially to halve peak FFTW memory
+    auto analytic_L = hilbertTransformSingle(left_signal);
+    // Left channel FFTW buffers are freed before right channel begins
+    auto analytic_R = hilbertTransformSingle(right_signal);
+
+    return {std::move(analytic_L), std::move(analytic_R)};
 }

src/complex_block.h

@@ -17,7 +17,12 @@
 
 class BlockHilbert {
 public:
-    std::pair<std::vector<std::complex<double>>, std::vector<std::complex<double>>> 
+    // Single-channel transform (lower peak memory)
+    std::vector<std::complex<double>>
+    hilbertTransformSingle(const std::vector<double>& signal);
+
+    // Stereo convenience (processes channels sequentially)
+    std::pair<std::vector<std::complex<double>>, std::vector<std::complex<double>>>
     hilbertTransform(const std::vector<double>& left_signal, const std::vector<double>& right_signal);
 };