[Bug] The numberOfFourierSamples may be changed after a window function is applied to the sound data in the menu command "View spectral slice“

[Chao Kong]

The sound editor has a menu command View spectral slice, which calls the Sound_and_Spectrum to do FFT. However, I noticed in the codes of this command, the selection was first applied to a window, and then passed to Sound_and_Spectrum with the fast argument set to true.

praat/foned/SoundAnalysisArea.cpp

Lines 913 to 935 in c3a783d

 static void CONVERT_DATA_TO_ONE__ViewSpectralSlice (SoundAnalysisArea me, EDITOR_ARGS) { 
 	CONVERT_DATA_TO_ONE 
 		const double start = ( my startSelection() == my endSelection() ? 
 			my instancePref_spectrogram_windowShape() == kSound_to_Spectrogram_windowShape::GAUSSIAN ? my startSelection() - my instancePref_spectrogram_windowLength() : 
 			my startSelection() - my instancePref_spectrogram_windowLength() / 2 : my startSelection() 
 		); 
 		const double finish = ( my startSelection() == my endSelection() ? 
 			my instancePref_spectrogram_windowShape() == kSound_to_Spectrogram_windowShape::GAUSSIAN ? my endSelection() + my instancePref_spectrogram_windowLength() : 
 			my endSelection() + my instancePref_spectrogram_windowLength() / 2 : my endSelection() 
 		); 
 		autoSound sound = extractSound (me, start, finish); 
 		Sound_multiplyByWindow (sound.get(), 
 			my instancePref_spectrogram_windowShape() == kSound_to_Spectrogram_windowShape::SQUARE ? kSound_windowShape::RECTANGULAR : 
 			my instancePref_spectrogram_windowShape() == kSound_to_Spectrogram_windowShape::HAMMING ? kSound_windowShape::HAMMING : 
 			my instancePref_spectrogram_windowShape() == kSound_to_Spectrogram_windowShape::BARTLETT ? kSound_windowShape::TRIANGULAR : 
 			my instancePref_spectrogram_windowShape() == kSound_to_Spectrogram_windowShape::WELCH ? kSound_windowShape::PARABOLIC : 
 			my instancePref_spectrogram_windowShape() == kSound_to_Spectrogram_windowShape::HANNING ? kSound_windowShape::HANNING : 
 			my instancePref_spectrogram_windowShape() == kSound_to_Spectrogram_windowShape::GAUSSIAN ? kSound_windowShape::GAUSSIAN_2 : kSound_windowShape::RECTANGULAR 
 		); 
 		autoSpectrum result = Sound_to_Spectrum (sound.get(), true); 
 	CONVERT_DATA_TO_ONE_END (Melder_cat (( my data() ? my data() -> name.get() : U"untitled" ), 
 			U"_", Melder_fixed (0.5 * (my startSelection() + my endSelection()), 3))) 
 } 

However, with fast set to true, the numberOfFourierSamples is likely to be changed:

praat/fon/Sound_and_Spectrum.cpp

Line 24 in c3a783d

 const integer numberOfFourierSamples = ( fast ? Melder_iroundUpToPowerOfTwo (my nx) : my nx ); 

This doesn't matter when the function is used to perform FFT to a whole sound, but in the case of the View spectral slice call, the data passed in has already been added to a window. In this situation, narrowing the numberOfFourierSamples is problematic.

The solution is setting the fast argument in this line to false to avoid narrowing the numberOfFourierSamples.

praat/foned/SoundAnalysisArea.cpp

Line 932 in c3a783d

 autoSpectrum result = Sound_to_Spectrum (sound.get(), true); 

This will not, in most cases, cause a speed concern, since the command View spectral slice is usually used to analyze a small segment of sounds.