People who want “authentic” sounds ahead of “good” sounds need read no further. ;-)
Recently I turned my attention to a particular note in a set of true recorded samples where SPEAR appeared to have let the motor noise get through. It was a D1 in a Roerfluit 8 set, and although SPEAR had been used to select only the wanted components, an unpleasant throbbing existed on that one note. It wasn’t loud, but it was unpleasant.
It occurred to me that what SPEAR had let through was not the motor noise itself, but rather the interaction of that noise with the fundamental component of the pipe sound. What we may have been hearing was normal “beating”, where one of the motor noise components (perhaps the main one) has a steady pitch very close to that of the pipe fundamental of about 73 Hz, and a similar level, at least at the location where the recording microphone had been placed.
SPEAR provided an ideal tool with which to test this theory, as well as surgically remove the fundamental and replace it with a prosthetic “good” one. It was a simple matter to remove the fundamental (Control “X”), and then save a D1sp.wav file containing the other harmonics. Then a new SPEAR file was created, to receive the fundamental from the clipboard. This was saved as a .wav file, with the name D1H1. In Audacity, it was possible to import both these files (onto two separate tracks). The D1H1 waveform showed clearly the presence of a beat: its amplitude undulated in a regular fashion.
What I then did was to use AdsynDX to create a new (synthesized) fundamental about 7 seconds in length, with a ramp attack of similar appearance to the D1H1. After retuning the D1sp.wav, and setting the new fundamental to a level determined by using the Audacity Analyzer on the D1H1, it was a simple matter to make a composite waveform of the two waveforms. We now had the original sound back, but without the throbbing. The new sample was soon looped, and was imported into the soundfont in place of the previous sample.
I find AdsynDX invaluable when making Short Hybrid Samples, as it allows the importing of CSV spreadsheet files which I use as a first stage in making the "accurate" synthesized loop sections. It has saved me many hours of work. It is not free, but its cost is very reasonable. However, one can make simple replacement waveforms such as that described in my post by using the free Klangsynth program, and then extending the one second sample to whatever length is required, by means of SPEAR.
Working today on some Short Hybrid Samples, I have been reminded that SPEAR itself can spoil the sound in some situations, without human error being the culprit. However, it was not difficult in one instance to use SPEAR to identify which harmonic it was getting wrong in its attack, and then to use the technique I described above to create a replacement harmonic which did not have the flaw which SPEAR had introduced.
Further to this thread which I began some months ago, I have encountered the noise problem again in a slightly different form, in the bottom notes of an Open Diapason 8 rank which I recently recorded. It belongs to a 12-stop Hill organ of 1870 imported to Sydney all that time ago. I recorded the whole organ, and I plan to release a jOrgan disposition/soundfont eventually. The problem made itself apparent not so much in the sound (although It can be heard if one listens carefully enough). What first puzzled me was that the waveform produced by SPEAR after selecting the wanted components was certainly not uniform across the 6 seconds of its duration. It “lurched” up and down somewhat during those 6 seconds, and this made the finding of good looping points much harder than it should have been. What is more, the basic waveform which was used for the SPEAR cleaning process showed nothing of that irregularity. This suggested to me that SPEAR itself was somehow responsible for the outcome.
I used SPEAR to isolate the first four harmonics of one of the notes in question, assigning each of them to a separate file and then examining them in turn using Audacity. The fundamental (i.e. the 1st Harmonic) showed considerable “unsteadiness”, although it did not seem to be extensive enough to cause what I have described. Anyway, I re-combined the four waveforms (very simple using Audacity), and the strange behaviour was there in the resulting waveform.
It seemed to me that what was happening is that the high level of organ noise level in these recordings, which would render them quite useless without the intervention of SPEAR, was causing SPEAR to misrepresent the phase of the fundamental over time. This was due to noise components being quite close in pitch to the fundamentals of these bottom notes. Although this gradually-changing phase difference was not in itself audible, it was certainly producing a change in the appearance of the waveform over the 6 seconds, and it is this which made the finding of good looping points more of a challenge.
The remedy is to create a reasonable “copy” of the fundamental but free of these discrepancies, as mentioned in my earlier post. This time I tried the simple and quick approach of copying the 2nd harmonic waveform to a new track in Audacity, halving its pitch (using Speed Adjust), adjusting its level appropriately, and using it as the new fundamental. This seems preferable to synthesizing a fundamental and then having to match its pitch accurately to the recorded sounds. It was, of course, necessary to combine this new fundamental with a waveform created by SPEAR from which the original fundamental had been removed. And as expected, the new full waveform was regular in shape over the 6 seconds, and could be easily looped.
That's an interesting aproach. Instead of synthesize a new fundamental you use the second fundamental transposed down one octave. I wonder if it would give te same result when you use the note without the bad fundamental line and transpose it in the sf or simply use a note close to it and transpose it only a little bit
Now you transpose over one octave, when using a close note the transposing distance is less.
What I have done is to remove the undesired part of the fundamental, or any other harmonic if neccesary, by magnifying the time scale until the dots (breakpoints) are visible. And then highlight the last dot of the initial part of the wave form and drag it onto exactly the top of the next breakpoint at the start of the next section checking that the frequency is the same throughtout. You may have to magnify the frequency scale also to do this.
Thank you for your interest in this thread. I apologize for not responding to your posts until now. Extensive radiotherapy for a potentially serious health condition has left me somewhat fatigued and preoccupied.
“Instead of synthesize a new fundamental you use the second fundamental transposed down one octave.”
Yes. But I would call it the “second harmonic”. I did it this way in an attempt to save time. Also, I thought that there would be fewer tuning complications. However, the second example I tried found the second harmonic also noise-infected. So the method was not a success in this case.
“I wonder if it would give te same result when you use the note without the bad fundamental line and transpose it in the sf”
No, to transpose such a note down an octave would simply produce a sound one octave lower, but it would be a sound with the fundamental missing.
“or simply use a note close to it and transpose it only a little bit”
In fact this is what I might choose to do in the “finishing” stage. My practice always has an eye on minimising the work for the creator, consistent with a high quality result. When making the samples, I always try to get the best result possible, within the time constraints I try to hold to. If in the finishing stage (regulating the notes at the console) I find a sample to be unsatisfactory (all things considered), then I will abandon it and use a sample close by. I will however avoid using the ADJACENT sample, as this will produce a zone (or “split”) of more than two notes. Regardless of whether I am using six or three samples per octave, I always try to maintain this standard: splits should never be more than two notes wide, as the discerning player will usually notice what is going on. I think that in most musical contexts, we can get away with it by using two-note splits. (This halves the work at one stroke. And if we limit ourselves to mono samples instead of stereo samples, we are down to a quarter of the work involved, and also a quarter of the soundfont size).
“Now you transpose over one octave, when using a close note the transposing distance is less.”
I don’t understand this statement.
“....... and drag it onto exactly the top of the next breakpoint at the start of the next section checking that the frequency is the same throughtout. You may have to magnify the frequency scale also to do this.”
I tried to follow your instructions, but I am not sure that I understood exactly what was involved (...“next breakpoint at the start of the next section” ??). Also, I had trouble dragging it enough for my 6-seconds sample length. Perhaps you could describe the process again in more detail? Your method seems a good idea (although the frequency aspect needs watching because of possible looping difficulties), but I think it needs to be something that one can achieve fairly quickly.
Another solution has just occurred to me. If for some reason a difficult sample really does have to be salvaged, and there appear to be tuning and therefore looping difficulties in matching a constructed fundamental (or other harmonic) to the main waveform, the constructed waveform could actually be made as a separate looped sample and given its own split in the soundfont Instrument , and so combined with the main sound during actual playing. It doesn’t seem very economical of the creator’s time, but it may provide a solution where for some reason it just has to be done.
I was able to download both versions of Chris' tutorial, but in each case I was not able to view it in any useful way. So I emailed Chris and he sent it to me as an attachment. The information there has enabled me to drag a harmonic of interest from an early breakpoint to the end of a 6-second sample (after deleting the rest of the harmonic). What I find however is that the waveform so produced by SPEAR is not level in amplitude along its duration, but actually reduces gradually towards zero at the end-point.
Has anyone else tried to do this with SPEAR, extending a harmonic over such a period of time? I remember trying this in the past, and getting the same result, and so abandoning the method, as the result was useless for my purposes. Is there some setting in SPEAR which prevents this behaviour?
Don't delete to the end but leave a couple of dots and lines at the end. When dragging or stretching spear is applying a kind of (cross)fading. When dragging completely to the end it will result in fade out.
I thought I had actually tried what Dries has suggested. From memory, what I did was to leave part of the fundamental line at the end, deleted the intervening noise-infected part, and stretched from near the start, joining it to the bit at the end. Perhaps I should not have joined it! ...Dries??
I tried it again, definitely joining it to a one-second portion left at the end of a seven-seconds waveform, and got the same diminishing amplitude result, followed by a slight gap before the one-second portion appears (as opened in Audacity as a .wav file). Definitely useless for looping purposes.
The problem is that when you drop the last dot of the left partial on the first dot of the right partial that you not have a real connection. You can do this by dropping it the last dot of the left partial just before the first dot of the right partial. Then select both partials and go in the menu to edit>join (Or Ctrl-J). The two partials will be joined together. When you now synthesize to sound the volume will not go down to zero at the junction point.
It is also possible to join the left and right partial without dragging.
Thank you. It works beautifully. I seem to recall that you got in touch with the author of SPEAR at some stage. This is not the first time you have shared some useful information about SPEAR and its functions.