view data/model/FFTModel.h @ 1196:c7b9c902642f spectrogram-minor-refactor

Fix threshold in spectrogram -- it wasn't working in the last release.
There is a new protocol for this. Formerly the threshold parameter had a
range from -50dB to 0 with the default at -50, and -50 treated internally
as "no threshold". However, there was a hardcoded, hidden internal threshold
for spectrogram colour mapping at -80dB with anything below this being rounded
to zero. Now the threshold parameter has range -81 to -1 with the default
at -80, -81 is treated internally as "no threshold", and there is no hidden
internal threshold. So the default behaviour is the same as before, an
effective -80dB threshold, but it is now possible to change this in both
directions. Sessions reloaded from prior versions may look slightly different
because, if the session says there should be no threshold, there will now
actually be no threshold instead of having the hidden internal one.

Still need to do something in the UI to make it apparent that the -81dB
setting removes the threshold entirely. This is at least no worse than the
previous, also obscured, magic -50dB setting.

/* -*- c-basic-offset: 4 indent-tabs-mode: nil -*-  vi:set ts=8 sts=4 sw=4: */

/*
    Sonic Visualiser
    An audio file viewer and annotation editor.
    Centre for Digital Music, Queen Mary, University of London.
    This file copyright 2006 Chris Cannam.
    
    This program is free software; you can redistribute it and/or
    modify it under the terms of the GNU General Public License as
    published by the Free Software Foundation; either version 2 of the
    License, or (at your option) any later version.  See the file
    COPYING included with this distribution for more information.
*/

#ifndef FFT_MODEL_H
#define FFT_MODEL_H

#include "DenseThreeDimensionalModel.h"
#include "DenseTimeValueModel.h"

#include "base/Window.h"

#include "data/fft/FFTapi.h"

#include <set>
#include <vector>
#include <complex>
#include <deque>

/**
 * An implementation of DenseThreeDimensionalModel that makes FFT data
 * derived from a DenseTimeValueModel available as a generic data
 * grid.
 */
class FFTModel : public DenseThreeDimensionalModel
{
    Q_OBJECT

    //!!! threading requirements?
    //!!! doubles? since we're not caching much

public:
    /**
     * Construct an FFT model derived from the given
     * DenseTimeValueModel, with the given window parameters and FFT
     * size (which may exceed the window size, for zero-padded FFTs).
     * 
     * If the model has multiple channels use only the given channel,
     * unless the channel is -1 in which case merge all available
     * channels.
     */
    FFTModel(const DenseTimeValueModel *model,
             int channel,
             WindowType windowType,
             int windowSize,
             int windowIncrement,
             int fftSize);
    ~FFTModel();

    // DenseThreeDimensionalModel and Model methods:
    //
    virtual int getWidth() const;
    virtual int getHeight() const;
    virtual float getValueAt(int x, int y) const { return getMagnitudeAt(x, y); }
    virtual bool isOK() const { return m_model && m_model->isOK(); }
    virtual sv_frame_t getStartFrame() const { return 0; }
    virtual sv_frame_t getEndFrame() const {
        return sv_frame_t(getWidth()) * getResolution() + getResolution();
    }
    virtual sv_samplerate_t getSampleRate() const {
        return isOK() ? m_model->getSampleRate() : 0;
    }
    virtual int getResolution() const { return m_windowIncrement; }
    virtual int getYBinCount() const { return getHeight(); }
    virtual float getMinimumLevel() const { return 0.f; } // Can't provide
    virtual float getMaximumLevel() const { return 1.f; } // Can't provide
    virtual Column getColumn(int x) const; // magnitudes
    virtual QString getBinName(int n) const;
    virtual bool shouldUseLogValueScale() const { return true; }
    virtual int getCompletion() const {
        int c = 100;
        if (m_model) {
            if (m_model->isReady(&c)) return 100;
        }
        return c;
    }
    virtual QString getError() const { return ""; } //!!!???
    virtual sv_frame_t getFillExtent() const { return getEndFrame(); }

    // FFTModel methods:
    //
    int getChannel() const { return m_channel; }
    WindowType getWindowType() const { return m_windowType; }
    int getWindowSize() const { return m_windowSize; }
    int getWindowIncrement() const { return m_windowIncrement; }
    int getFFTSize() const { return m_fftSize; }
    
    float getMagnitudeAt(int x, int y) const;
    float getMaximumMagnitudeAt(int x) const;
    float getPhaseAt(int x, int y) const;
    void getValuesAt(int x, int y, float &real, float &imaginary) const;
    bool getMagnitudesAt(int x, float *values, int minbin = 0, int count = 0) const;
    bool getPhasesAt(int x, float *values, int minbin = 0, int count = 0) const;
    bool getValuesAt(int x, float *reals, float *imaginaries, int minbin = 0, int count = 0) const;

    /**
     * Calculate an estimated frequency for a stable signal in this
     * bin, using phase unwrapping.  This will be completely wrong if
     * the signal is not stable here.
     */
    virtual bool estimateStableFrequency(int x, int y, double &frequency);

    enum PeakPickType
    {
        AllPeaks,                /// Any bin exceeding its immediate neighbours
        MajorPeaks,              /// Peaks picked using sliding median window
        MajorPitchAdaptivePeaks  /// Bigger window for higher frequencies
    };

    typedef std::set<int> PeakLocationSet; // bin
    typedef std::map<int, double> PeakSet; // bin -> freq

    /**
     * Return locations of peak bins in the range [ymin,ymax].  If
     * ymax is zero, getHeight()-1 will be used.
     */
    virtual PeakLocationSet getPeaks(PeakPickType type, int x,
                                     int ymin = 0, int ymax = 0) const;

    /**
     * Return locations and estimated stable frequencies of peak bins.
     */
    virtual PeakSet getPeakFrequencies(PeakPickType type, int x,
                                       int ymin = 0, int ymax = 0) const;

    QString getTypeName() const { return tr("FFT"); }

public slots:
    void sourceModelAboutToBeDeleted();

private:
    FFTModel(const FFTModel &); // not implemented
    FFTModel &operator=(const FFTModel &); // not implemented

    const DenseTimeValueModel *m_model;
    int m_channel;
    WindowType m_windowType;
    int m_windowSize;
    int m_windowIncrement;
    int m_fftSize;
    Window<float> m_windower;
    FFTForward m_fft;
    
    int getPeakPickWindowSize(PeakPickType type, sv_samplerate_t sampleRate,
                              int bin, float &percentile) const;

    std::pair<sv_frame_t, sv_frame_t> getSourceSampleRange(int column) const {
        sv_frame_t startFrame = m_windowIncrement * sv_frame_t(column);
        sv_frame_t endFrame = startFrame + m_windowSize;
        // Cols are centred on the audio sample (e.g. col 0 is centred at sample 0)
        startFrame -= m_windowSize / 2;
        endFrame -= m_windowSize / 2;
        return { startFrame, endFrame };
    }

    std::vector<std::complex<float> > getFFTColumn(int column) const;
    std::vector<float> getSourceSamples(int column) const;
    std::vector<float> getSourceData(std::pair<sv_frame_t, sv_frame_t>) const;
    std::vector<float> getSourceDataUncached(std::pair<sv_frame_t, sv_frame_t>) const;

    struct SavedSourceData {
        std::pair<sv_frame_t, sv_frame_t> range;
        std::vector<float> data;
    };
    mutable SavedSourceData m_savedData;
    
    struct SavedColumn {
        int n;
        std::vector<std::complex<float> > col;
    };
    mutable std::deque<SavedColumn> m_cached;
    size_t m_cacheSize;
};

#endif