The following examples were designed to show the cumulative effects
of the applied techniques
MIDI performance, with adjusted amplitudes and timing
(This example was generated by a commercial sample-based synthesizer,
driven by a MIDI representation of the score. Onset times and amplitudes
were measured from the "real" performance, and the onset times and amplitudes
of the MIDI notes were adjusted correspondingly.)
Single wavetable oscillator + real amplitude curve
(At this example, a single wavetable oscillator generates the sound
throughout the whole phrase. The stored wavetable was created from the
spectrum of the C5 note of the "real" performance, measured at its peak
amplitude. The amplitude fluctuation of the synthesized sound is realized
using the measured amplitude curve of the "real" performance. The pitches
of the notes are constant values.)
Single wavetable oscillator + real amplitude and frequency curves
(Same setup as with the previous example, except the frequency fluctuation
of the synthesized sound is also realized using the measured pitch curve
of the "real" performance.)
Multiple wavetable oscillator + real amplitude and frequency curves
(At this example, a different wavetable is used for each different
pitches in the phrase, otherwise, the setup is the same as with the previous
example. The stored wavetable for each pitch was created from the spectrum
of the corresponding note of the "real" performance, measured at its peak
amplitude.)
CSIS Instrument Model performance, using measured amplitude and frequency
curves
(At this example, the Instrument Model of the CSIS Model is driven
by amplitude and frequency curves, which were measured from the "real"
performance.)
CSIS Instrument Model performance, using measured amplitude and frequency
curves + sampled attack
(Same setup as with the previous example, augmented with the attack
sample splicing technique)
CSIS Instrument Model performance, Single Envelope
(This example was designed to illustrate that some simple approach
to create the control signals does not result in an acceptable performance.
The amplitude curve of this example was created by measuring the amplitude
curve of the first note in the "real" performance, and splicing together
as many instances of that as many notes in the phrase are. Even more, the
amplitude values between slurred notes in the resulted curve were increased
from zero to a relatively high level, to increase fidelity. Still, the
resulted performance is not satisfying.)
Complete CSIS (Performance + Instrument) Model performance