1 /***************************************************************************
2 * Copyright (C) 2010 by Simon Andreas Eugster (simon.eu@gmail.com) *
3 * This file is part of kdenlive. See www.kdenlive.org. *
5 * This program is free software; you can redistribute it and/or modify *
6 * it under the terms of the GNU General Public License as published by *
7 * the Free Software Foundation; either version 2 of the License, or *
8 * (at your option) any later version. *
9 ***************************************************************************/
13 #include "audiospectrum.h"
15 #include "tools/kiss_fftr.h"
19 #include <QMouseEvent>
23 // Enables debugging, like writing a GNU Octave .m file to /tmp
24 //#define DEBUG_AUDIOSPEC
25 #ifdef DEBUG_AUDIOSPEC
27 bool fileWritten = false;
30 #define MIN_DB_VALUE -120
32 const QString AudioSpectrum::directions[] = {"North", "Northeast", "East", "Southeast"};
34 AudioSpectrum::AudioSpectrum(QWidget *parent) :
35 AbstractAudioScopeWidget(false, parent),
38 m_rescaleVerticalThreshold(2.0f),
39 m_rescaleActive(false),
40 m_rescalePropertiesLocked(false),
43 ui = new Ui::AudioSpectrum_UI;
46 m_distance = QSize(65, 30);
50 m_aLockHz = new QAction(i18n("Lock maximum frequency"), this);
51 m_aLockHz->setCheckable(true);
52 m_aLockHz->setEnabled(false);
55 m_menu->addSeparator();
56 m_menu->addAction(m_aLockHz);
59 ui->windowSize->addItem("256", QVariant(256));
60 ui->windowSize->addItem("512", QVariant(512));
61 ui->windowSize->addItem("1024", QVariant(1024));
62 ui->windowSize->addItem("2048", QVariant(2048));
64 ui->windowFunction->addItem(i18n("Rectangular window"), FFTTools::Window_Rect);
65 ui->windowFunction->addItem(i18n("Triangular window"), FFTTools::Window_Triangle);
66 ui->windowFunction->addItem(i18n("Hamming window"), FFTTools::Window_Hamming);
69 m_cfg = kiss_fftr_alloc(ui->windowSize->itemData(ui->windowSize->currentIndex()).toInt(), 0,0,0);
70 //m_windowFunctions.insert("tri512", FFTTools::window(FFTTools::Window_Hamming, 8, 0));
71 // TODO Window function cache
75 b &= connect(ui->windowSize, SIGNAL(currentIndexChanged(int)), this, SLOT(slotUpdateCfg()));
78 AbstractScopeWidget::init();
80 AudioSpectrum::~AudioSpectrum()
88 void AudioSpectrum::readConfig()
90 AbstractScopeWidget::readConfig();
92 KSharedConfigPtr config = KGlobal::config();
93 KConfigGroup scopeConfig(config, AbstractScopeWidget::configName());
94 m_aLockHz->setChecked(scopeConfig.readEntry("lockHz", false));
95 ui->windowSize->setCurrentIndex(scopeConfig.readEntry("windowSize", 0));
96 m_dBmax = scopeConfig.readEntry("dBmax", 0);
97 m_dBmin = scopeConfig.readEntry("dBmin", -70);
98 ui->windowFunction->setCurrentIndex(scopeConfig.readEntry("windowFunction", 0));
100 void AudioSpectrum::writeConfig()
102 KSharedConfigPtr config = KGlobal::config();
103 KConfigGroup scopeConfig(config, AbstractScopeWidget::configName());
104 scopeConfig.writeEntry("windowSize", ui->windowSize->currentIndex());
105 scopeConfig.writeEntry("windowFunction", ui->windowFunction->currentIndex());
106 scopeConfig.writeEntry("lockHz", m_aLockHz->isChecked());
107 scopeConfig.writeEntry("dBmax", m_dBmax);
108 scopeConfig.writeEntry("dBmin", m_dBmin);
112 QString AudioSpectrum::widgetName() const { return QString("AudioSpectrum"); }
113 bool AudioSpectrum::isBackgroundDependingOnInput() const { return false; }
114 bool AudioSpectrum::isScopeDependingOnInput() const { return true; }
115 bool AudioSpectrum::isHUDDependingOnInput() const { return false; }
117 QImage AudioSpectrum::renderBackground(uint) { return QImage(); }
119 QImage AudioSpectrum::renderAudioScope(uint, const QVector<int16_t> audioFrame, const int freq, const int num_channels, const int num_samples)
121 if (audioFrame.size() > 63) {
122 m_freqMax = freq / 2;
124 QTime start = QTime::currentTime();
126 bool customCfg = false;
127 kiss_fftr_cfg myCfg = m_cfg;
128 int fftWindow = ui->windowSize->itemData(ui->windowSize->currentIndex()).toInt();
129 if (fftWindow > num_samples) {
130 fftWindow = num_samples;
133 if ((fftWindow & 1) == 1) {
138 myCfg = kiss_fftr_alloc(fftWindow, 0,0,0);
141 float data[fftWindow];
142 float freqSpectrum[fftWindow/2];
145 for (int i = 0; i < fftWindow; i++) {
146 if (audioFrame.data()[i*num_channels] > maxSig) {
147 maxSig = audioFrame.data()[i*num_channels];
151 // Prepare frequency space vector. The resulting FFT vector is only half as long.
152 kiss_fft_cpx freqData[fftWindow/2];
156 // Copy the first channel's audio into a vector for the FFT display
157 // (only one channel handled at the moment)
158 if (num_samples < fftWindow) {
159 std::fill(&data[num_samples], &data[fftWindow-1], 0);
162 FFTTools::WindowType windowType = (FFTTools::WindowType) ui->windowFunction->itemData(ui->windowFunction->currentIndex()).toInt();
163 QVector<float> window;
164 float windowScaleFactor = 1;
165 if (windowType != FFTTools::Window_Rect) {
166 window = FFTTools::window(windowType, fftWindow, 0);
167 windowScaleFactor = 1.0/window[fftWindow];
168 qDebug() << "Using a window scaling factor of " << windowScaleFactor;
171 // Normalize signals to [0,1] to get correct dB values later on
172 for (int i = 0; i < num_samples && i < fftWindow; i++) {
173 if (windowType != FFTTools::Window_Rect) {
174 data[i] = (float) audioFrame.data()[i*num_channels] / 32767.0f * window[i];
176 data[i] = (float) audioFrame.data()[i*num_channels] / 32767.0f;
180 // Calculate the Fast Fourier Transform for the input data
181 kiss_fftr(myCfg, data, freqData);
184 // Logarithmic scale: 20 * log ( 2 * magnitude / N ) with magnitude = sqrt(r² + i²)
185 // with N = FFT size (after FFT, 1/2 window size)
186 for (int i = 0; i < fftWindow/2; i++) {
187 // Logarithmic scale: 20 * log ( 2 * magnitude / N ) with magnitude = sqrt(r² + i²)
188 // with N = FFT size (after FFT, 1/2 window size)
189 freqSpectrum[i] = 20*log(pow(pow(fabs(freqData[i].r * windowScaleFactor),2) + pow(fabs(freqData[i].i * windowScaleFactor),2), .5)/((float)fftWindow/2.0f))/log(10);;
195 QImage spectrum(m_scopeRect.size(), QImage::Format_ARGB32);
196 spectrum.fill(qRgba(0,0,0,0));
197 const uint w = m_innerScopeRect.width();
198 const uint h = m_innerScopeRect.height();
199 const uint leftDist = m_innerScopeRect.left() - m_scopeRect.left();
200 const uint topDist = m_innerScopeRect.top() - m_scopeRect.top();
205 for (uint i = 0; i < w; i++) {
207 // i: Pixel coordinate
208 // x: Frequency array index (float!) corresponding to the pixel
211 x = i/((float) w) * fftWindow/2;
214 // Use linear interpolation in order to get smoother display
215 if (i == 0 || i == w-1) {
216 val = freqSpectrum[i];
219 if (freqSpectrum[xi] > freqSpectrum[xi+1]
221 // This is a hack to preserve peaks.
222 // Consider f = {0, 100, 0}
224 // Then x is 50 both times, and the 100 peak is lost.
225 // Get it back here for the first x after the peak.
226 val = freqSpectrum[xi];
228 val = (xi+1 - x) * freqSpectrum[xi]
229 + (x - xi) * freqSpectrum[xi+1];
233 // freqSpectrum values range from 0 to -inf as they are relative dB values.
234 for (uint y = 0; y < h*(1 - (val - m_dBmax)/(m_dBmin-m_dBmax)) && y < h; y++) {
235 spectrum.setPixel(leftDist + i, topDist + h-y-1, qRgba(225, 182, 255, 255));
241 emit signalScopeRenderingFinished(start.elapsed(), 1);
243 #ifdef DEBUG_AUDIOSPEC
244 if (!fileWritten || true) {
246 mFile.open("/tmp/freq.m");
248 qDebug() << "Opening file failed.";
252 for (int sample = 0; sample < 256; sample++) {
253 mFile << data[sample] << " ";
257 mFile << "freq = [ ";
258 for (int sample = 0; sample < 256; sample++) {
259 mFile << freqData[sample].r << "+" << freqData[sample].i << "*i ";
265 qDebug() << "File written.";
268 qDebug() << "File already written.";
278 emit signalScopeRenderingFinished(0, 1);
282 QImage AudioSpectrum::renderHUD(uint)
284 QTime start = QTime::currentTime();
286 // Minimum distance between two lines
287 const uint minDistY = 30;
288 const uint minDistX = 40;
289 const uint textDistX = 10;
290 const uint textDistY = 25;
291 const uint topDist = m_innerScopeRect.top() - m_scopeRect.top();
292 const uint leftDist = m_innerScopeRect.left() - m_scopeRect.left();
293 const uint dbDiff = ceil((float)minDistY/m_innerScopeRect.height() * (m_dBmax-m_dBmin));
295 QImage hud(m_scopeRect.size(), QImage::Format_ARGB32);
296 hud.fill(qRgba(0,0,0,0));
298 QPainter davinci(&hud);
299 davinci.setPen(AbstractAudioScopeWidget::penLight);
302 for (int db = -dbDiff; db > m_dBmin; db -= dbDiff) {
303 y = topDist + m_innerScopeRect.height() * ((float)db)/(m_dBmin - m_dBmax);
304 if (y-topDist > m_innerScopeRect.height()-minDistY+10) {
305 // Abort here, there is still a line left for min dB to paint which needs some room.
308 davinci.drawLine(leftDist, y, leftDist + m_innerScopeRect.width()-1, y);
309 davinci.drawText(leftDist + m_innerScopeRect.width() + textDistX, y + 6, i18n("%1 dB", m_dBmax + db));
311 davinci.drawLine(leftDist, topDist, leftDist + m_innerScopeRect.width()-1, topDist);
312 davinci.drawText(leftDist + m_innerScopeRect.width() + textDistX, topDist+6, i18n("%1 dB", m_dBmax));
313 davinci.drawLine(leftDist, topDist+m_innerScopeRect.height()-1, leftDist + m_innerScopeRect.width()-1, topDist+m_innerScopeRect.height()-1);
314 davinci.drawText(leftDist + m_innerScopeRect.width() + textDistX, topDist+m_innerScopeRect.height()+6, i18n("%1 dB", m_dBmin));
316 // TODO max Hz dynamically
317 const uint hzDiff = ceil( ((float)minDistX)/m_innerScopeRect.width() * m_freqMax / 1000 ) * 1000;
319 y = topDist + m_innerScopeRect.height() + textDistY;
320 for (uint hz = 0; hz <= m_freqMax; hz += hzDiff) {
321 x = leftDist + m_innerScopeRect.width() * ((float)hz)/m_freqMax;
322 davinci.drawLine(x, topDist, x, topDist + m_innerScopeRect.height()+6);
323 if (hz < m_freqMax) {
324 davinci.drawText(x-4, y, QVariant(hz/1000).toString());
326 davinci.drawText(x-10, y, i18n("%1 kHz",hz/1000));
330 for (uint dHz = 1; dHz < 4; dHz++) {
331 x = leftDist + m_innerScopeRect.width() * ((float)hz - dHz * hzDiff/4.0f)/m_freqMax;
332 davinci.drawLine(x, topDist, x, topDist + m_innerScopeRect.height()-1);
338 emit signalHUDRenderingFinished(start.elapsed(), 1);
342 QRect AudioSpectrum::scopeRect() {
346 ui->verticalSpacer->geometry().top()+6 // Top
348 AbstractAudioScopeWidget::rect().bottomRight()
350 m_innerScopeRect = QRect(
352 m_scopeRect.left()+6, // Left
353 m_scopeRect.top()+6 // Top
355 ui->verticalSpacer->geometry().right()-70,
356 ui->verticalSpacer->geometry().bottom()-40
363 void AudioSpectrum::slotUpdateCfg()
366 m_cfg = kiss_fftr_alloc(ui->windowSize->itemData(ui->windowSize->currentIndex()).toInt(), 0,0,0);
372 void AudioSpectrum::mouseMoveEvent(QMouseEvent *event)
374 QPoint movement = event->pos()-m_rescaleStartPoint;
376 if (m_rescaleActive) {
377 if (m_rescalePropertiesLocked) {
378 // Direction is known, now adjust parameters
380 // Reset the starting point to make the next moveEvent relative to the current one
381 m_rescaleStartPoint = event->pos();
384 if (!m_rescaleFirstRescaleDone) {
385 // We have just learned the desired direction; Normalize the movement to one pixel
386 // to avoid a jump by m_rescaleMinDist
388 if (movement.x() != 0) {
389 movement.setX(movement.x() / abs(movement.x()));
391 if (movement.y() != 0) {
392 movement.setY(movement.y() / abs(movement.y()));
395 m_rescaleFirstRescaleDone = true;
398 if (m_rescaleClockDirection == AudioSpectrum::North) {
399 // Nort-South direction: Adjust the dB scale
401 if ((m_rescaleModifiers & Qt::ShiftModifier) == 0) {
403 // By default adjust the min dB value
404 m_dBmin += movement.y();
408 // Adjust max dB value if Shift is pressed.
409 m_dBmax += movement.y();
413 // Ensure the dB values lie in [-100, 0] (or rather [MIN_DB_VALUE, 0])
414 // 0 is the upper bound, everything below -70 dB is most likely noise
418 if (m_dBmin < MIN_DB_VALUE) {
419 m_dBmin = MIN_DB_VALUE;
421 // Ensure there is at least 6 dB between the minimum and the maximum value;
422 // lower values hardly make sense
423 if (m_dBmax - m_dBmin < 6) {
424 if ((m_rescaleModifiers & Qt::ShiftModifier) == 0) {
425 // min was adjusted; Try to adjust the max value to maintain the
426 // minimum dB difference of 6 dB
427 m_dBmax = m_dBmin + 6;
433 // max was adjusted, adjust min
434 m_dBmin = m_dBmax - 6;
435 if (m_dBmin < MIN_DB_VALUE) {
436 m_dBmin = MIN_DB_VALUE;
437 m_dBmax = MIN_DB_VALUE+6;
449 // Detect the movement direction here.
450 // This algorithm relies on the aspect ratio of dy/dx (size and signum).
451 if (movement.manhattanLength() > m_rescaleMinDist) {
452 float diff = ((float) movement.y())/movement.x();
454 if (abs(diff) > m_rescaleVerticalThreshold || movement.x() == 0) {
455 m_rescaleClockDirection = AudioSpectrum::North;
456 } else if (abs(diff) < 1/m_rescaleVerticalThreshold) {
457 m_rescaleClockDirection = AudioSpectrum::East;
458 } else if (diff < 0) {
459 m_rescaleClockDirection = AudioSpectrum::Northeast;
461 m_rescaleClockDirection = AudioSpectrum::Southeast;
463 #ifdef DEBUG_AUDIOSPEC
464 qDebug() << "Diff is " << diff << "; chose " << directions[m_rescaleClockDirection] << " as direction";
466 m_rescalePropertiesLocked = true;
470 AbstractAudioScopeWidget::mouseMoveEvent(event);
474 void AudioSpectrum::mousePressEvent(QMouseEvent *event)
476 if (event->button() == Qt::LeftButton) {
477 // Rescaling mode starts
478 m_rescaleActive = true;
479 m_rescalePropertiesLocked = false;
480 m_rescaleFirstRescaleDone = false;
481 m_rescaleStartPoint = event->pos();
482 m_rescaleModifiers = event->modifiers();
485 AbstractAudioScopeWidget::mousePressEvent(event);
489 void AudioSpectrum::mouseReleaseEvent(QMouseEvent *event)
491 m_rescaleActive = false;
492 m_rescalePropertiesLocked = false;
494 AbstractAudioScopeWidget::mouseReleaseEvent(event);
498 #ifdef DEBUG_AUDIOSPEC
499 #undef DEBUG_AUDIOSPEC