Pipe Organ tones without samples
At present, all digital Pipe Organs on the market*, be they applications running on computers and devices, or as physical organ instruments, synthesize their sounds in one of two ways :
SAMPLE REPLAY SYNTHESIS
This is the most common synthesis technique for Virtual Pipe Organs, and is the easiest to understand. The organ developers make a sound recording of every note, for every rank of pipes in the organ. If a Division of the organ contains a Tremulant, then all the ranks that can undergo Tremulation are recorded twice, with and without the effect. This results in many thousands of audio files for even a small organ. When the Virtual Organ is played, the appropriate set of recordings is selected, and played back by the application. So at any moment a large number of recordings are being played back - one for every pipe currently speaking, which can be many hundreds if a large number of stops are pulled out. This approach to Virtual Pipe Organ design leads to an extremely realistic and compelling overall sound, but suffers from two problems - titanic storage requirements, and a 'printed reverberation' problem. The storage problem is easy to understand - all those recordings, several seconds long, rapidly add up to at minimum hundreds of MBytes, usually multiple GigaBytes of storage space. The reverb problem is more subtle. The organ that was recorded to make the sample set is located typically within a church, and churches are very reverberant spaces. Of course, the reverberant space is one of the major factors in making church organs sound so great, but the sampling process inevitably captures the sound of the reverberant space where the organ is being played. So at the point you are playing back the Virtual Organ, no matter how small your church, or your rehearsal room, if the samples you are playing were recorded in a huge Cathedral, that will be the sound you get. Which means you will struggle to get a nice 'Continuo Organ' sound, with the merest hint of reverberation. Plus of course if you are playing in a Church, your own church will add reverberation on top of the recorded reverberation, potentially making the resulting sound very muddy, and just wrong.
MODELED SYNTHESIS
This technique is the more modern way of implementing a Virtual Pipe Organ. Instead of storing recordings of pipes, an approximate mathematical model of the pipe is devised. This mathematical model drives a real-time simulation when the organ is played, to generate the 'correct' sounds of the pipes. This can lead to quite beautiful synthesis, it is very flexible as it opens the Virtual Organ up to personal custom voicing, and it solves the reverberation problem completely, allowing you to either use the natural reverberation of your church, or an artificial reverberator to add a bit of space to your sounds. But a modeled organ can only ever be an approximation, as the numerical model of the pipe has to be simplified in order to make the computation not overload the host computer. With these simplifications come shortcomings in the simulated tones, yet despite the approximations and shortcuts, a modelled organ is still VERY CPU-intensive. Even the most powerful computers will hiccup and fail when faced with playing a quality modelled organ with all stops out.
THE THIRD WAY - HARMONIC ANALYSIS
We chose to do things differently when we built the synthesis engine for our iOS and Mac OS organs. Our technique starts with samples, but instead of playing them back directly we analyse the audio to create a set of "harmonic fingerprints" for each pipe, representing different points in time during a played note. So, rather than analysing the pipes and building a numerical model, we analyse the sounds generated by the pipes, and build a very compact representation of those sounds. We compute 16 different fingerprints for each pipe, 8 dedicated to the earliest 'attack phase' of the note, 8 dedicated to the sustain phase, and convert these fingerprints into wavetables as the app is launched. A wavetable is effectively a microsample of a single cycle of the waveform of a note, stored as 1024 16-bit values. When you play the organ, Church Organ charts a trajectory through these wavetables to synthesize its audio. Our playback is as efficient as sample replay synthesis - your Mac will never run out of performance playing the organ, a distinct possibility on CPU-heavy "modelled" organs - but it requires only a tiny fraction of the storage of a sample-based organ, and our storage requirement is more like that of modeled synthesis.
Our synthesis is also naturally free of any reverberation that was present in the original samples - the 'giant Cathedral reverberation' that can make sampled organs sound so out of place in a small space is absent, and you are then free to dial in as much or as little of our artificial reverberation engine as you wish, or you can just use the natural acoustics of your performance space as a reverberation chamber. So, just like a modeled organ, our Church Organ sounds perfectly authentic inside small churches and rehearsal / performance halls. Also, our signals are guaranteed to be free of noise, because only harmonic information is represented in the wavetables, so our signals are beautifully clean, with no keyclicks or blower noise.
We like to think that this synthesis technique gives us the best of all worlds - very small download size, very low CPU requirement, tiny run-time memory footprint, and to our ears an authentic, lovely-sounding set of Pipe Organ tones. You may find the absence of keyclicks, blower and other mechanical noises initially disconcerting - give our audio samples and videos a listen, and judge for yourselves.
*of course this is a rather sweeping statement - let us know if we're wrong!
SAMPLE REPLAY SYNTHESIS
This is the most common synthesis technique for Virtual Pipe Organs, and is the easiest to understand. The organ developers make a sound recording of every note, for every rank of pipes in the organ. If a Division of the organ contains a Tremulant, then all the ranks that can undergo Tremulation are recorded twice, with and without the effect. This results in many thousands of audio files for even a small organ. When the Virtual Organ is played, the appropriate set of recordings is selected, and played back by the application. So at any moment a large number of recordings are being played back - one for every pipe currently speaking, which can be many hundreds if a large number of stops are pulled out. This approach to Virtual Pipe Organ design leads to an extremely realistic and compelling overall sound, but suffers from two problems - titanic storage requirements, and a 'printed reverberation' problem. The storage problem is easy to understand - all those recordings, several seconds long, rapidly add up to at minimum hundreds of MBytes, usually multiple GigaBytes of storage space. The reverb problem is more subtle. The organ that was recorded to make the sample set is located typically within a church, and churches are very reverberant spaces. Of course, the reverberant space is one of the major factors in making church organs sound so great, but the sampling process inevitably captures the sound of the reverberant space where the organ is being played. So at the point you are playing back the Virtual Organ, no matter how small your church, or your rehearsal room, if the samples you are playing were recorded in a huge Cathedral, that will be the sound you get. Which means you will struggle to get a nice 'Continuo Organ' sound, with the merest hint of reverberation. Plus of course if you are playing in a Church, your own church will add reverberation on top of the recorded reverberation, potentially making the resulting sound very muddy, and just wrong.
MODELED SYNTHESIS
This technique is the more modern way of implementing a Virtual Pipe Organ. Instead of storing recordings of pipes, an approximate mathematical model of the pipe is devised. This mathematical model drives a real-time simulation when the organ is played, to generate the 'correct' sounds of the pipes. This can lead to quite beautiful synthesis, it is very flexible as it opens the Virtual Organ up to personal custom voicing, and it solves the reverberation problem completely, allowing you to either use the natural reverberation of your church, or an artificial reverberator to add a bit of space to your sounds. But a modeled organ can only ever be an approximation, as the numerical model of the pipe has to be simplified in order to make the computation not overload the host computer. With these simplifications come shortcomings in the simulated tones, yet despite the approximations and shortcuts, a modelled organ is still VERY CPU-intensive. Even the most powerful computers will hiccup and fail when faced with playing a quality modelled organ with all stops out.
THE THIRD WAY - HARMONIC ANALYSIS
We chose to do things differently when we built the synthesis engine for our iOS and Mac OS organs. Our technique starts with samples, but instead of playing them back directly we analyse the audio to create a set of "harmonic fingerprints" for each pipe, representing different points in time during a played note. So, rather than analysing the pipes and building a numerical model, we analyse the sounds generated by the pipes, and build a very compact representation of those sounds. We compute 16 different fingerprints for each pipe, 8 dedicated to the earliest 'attack phase' of the note, 8 dedicated to the sustain phase, and convert these fingerprints into wavetables as the app is launched. A wavetable is effectively a microsample of a single cycle of the waveform of a note, stored as 1024 16-bit values. When you play the organ, Church Organ charts a trajectory through these wavetables to synthesize its audio. Our playback is as efficient as sample replay synthesis - your Mac will never run out of performance playing the organ, a distinct possibility on CPU-heavy "modelled" organs - but it requires only a tiny fraction of the storage of a sample-based organ, and our storage requirement is more like that of modeled synthesis.
Our synthesis is also naturally free of any reverberation that was present in the original samples - the 'giant Cathedral reverberation' that can make sampled organs sound so out of place in a small space is absent, and you are then free to dial in as much or as little of our artificial reverberation engine as you wish, or you can just use the natural acoustics of your performance space as a reverberation chamber. So, just like a modeled organ, our Church Organ sounds perfectly authentic inside small churches and rehearsal / performance halls. Also, our signals are guaranteed to be free of noise, because only harmonic information is represented in the wavetables, so our signals are beautifully clean, with no keyclicks or blower noise.
We like to think that this synthesis technique gives us the best of all worlds - very small download size, very low CPU requirement, tiny run-time memory footprint, and to our ears an authentic, lovely-sounding set of Pipe Organ tones. You may find the absence of keyclicks, blower and other mechanical noises initially disconcerting - give our audio samples and videos a listen, and judge for yourselves.
*of course this is a rather sweeping statement - let us know if we're wrong!