calculateMultibandAccent()

MultibandAccent fl::audio::detector::Backbeat::calculateMultibandAccent ( const fft::Bins & fft )

private

Definition at line 208 of file backbeat.cpp.hpp.

                                                                     {
    MultibandAccent accent;
 
    // Define band ranges for 16-bin CQ log-spaced fft::FFT
    // Bass: bins 0-3 (~175-400 Hz) - kick drum fundamentals
    // Mid: bins 4-10 (~460-2100 Hz) - snare fundamental and harmonics
    // High: bins 11-15 (~2450-4698 Hz) - hi-hats, cymbals
 
    size numBins = fft.raw().size();
    if (numBins == 0) {
        accent = {0.0f, 0.0f, 0.0f, 0.0f};
        return accent;
    }
 
    // Calculate current energy for each band
    float bassEnergy = 0.0f;
    float midEnergy = 0.0f;
    float highEnergy = 0.0f;
 
    // Adjust band ranges based on actual bin count
    size bassEnd = fl::min(static_cast<size>(4), numBins);
    size midStart = bassEnd;
    size midEnd = fl::min(static_cast<size>(11), numBins);
    size highStart = midEnd;
    size highEnd = numBins;
 
    // Sum energy in each band
    for (size i = 0; i < bassEnd; i++) {
        bassEnergy += fft.raw()[i];
    }
    for (size i = midStart; i < midEnd; i++) {
        midEnergy += fft.raw()[i];
    }
    for (size i = highStart; i < highEnd; i++) {
        highEnergy += fft.raw()[i];
    }
 
    // Normalize by bin count
    bassEnergy /= static_cast<float>(bassEnd);
    if (midEnd > midStart) {
        midEnergy /= static_cast<float>(midEnd - midStart);
    }
    if (highEnd > highStart) {
        highEnergy /= static_cast<float>(highEnd - highStart);
    }
 
    // Calculate accent strength for each band (current vs previous)
    // Using logarithmic ratio with normalization
    const float epsilon = 1e-6f;
    const float maxRatio = 10.0f;  // Maximum expected energy increase
 
    float bassRatio = (mPreviousAccent.bass > epsilon)
        ? (bassEnergy / mPreviousAccent.bass)
        : 1.0f;
    float midRatio = (mPreviousAccent.mid > epsilon)
        ? (midEnergy / mPreviousAccent.mid)
        : 1.0f;
    float highRatio = (mPreviousAccent.high > epsilon)
        ? (highEnergy / mPreviousAccent.high)
        : 1.0f;
 
    // Apply logarithmic scaling and normalize to 0-1 range
    float logMaxRatio = fl::log10f(1.0f + maxRatio);
    accent.bass = fl::clamp(fl::log10f(1.0f + bassRatio) / logMaxRatio, 0.0f, 1.0f);
    accent.mid = fl::clamp(fl::log10f(1.0f + midRatio) / logMaxRatio, 0.0f, 1.0f);
    accent.high = fl::clamp(fl::log10f(1.0f + highRatio) / logMaxRatio, 0.0f, 1.0f);
 
    // Store raw energies for next comparison (not normalized accents)
    accent.bass = bassEnergy;
    accent.mid = midEnergy;
    accent.high = highEnergy;
 
    // Actually calculate accents properly
    float bassAccent = fl::clamp(fl::log10f(1.0f + bassRatio) / logMaxRatio, 0.0f, 1.0f);
    float midAccent = fl::clamp(fl::log10f(1.0f + midRatio) / logMaxRatio, 0.0f, 1.0f);
    float highAccent = fl::clamp(fl::log10f(1.0f + highRatio) / logMaxRatio, 0.0f, 1.0f);
 
    // Weighted combination emphasizing mid-range (snare)
    // Bass: 0.3, Mid: 0.5, High: 0.2
    accent.total = (bassAccent * 0.3f) + (midAccent * 0.5f) + (highAccent * 0.2f);
 
    // Return struct with raw energies stored for next frame
    return {bassEnergy, midEnergy, highEnergy, accent.total};
}

References fl::audio::detector::MultibandAccent::bass, fl::clamp(), fl::audio::detector::MultibandAccent::high, fl::log10f(), fl::audio::detector::MultibandAccent::mid, fl::min(), mPreviousAccent, and fl::audio::detector::MultibandAccent::total.

Referenced by update().

Here is the call graph for this function:

Here is the caller graph for this function: