Additive Synthesis is a
powerful technique
|
Additive synthesis is a powerful technique
that gives access to the very fine structure of sound. In fact, every partial
can be controlled independantly which gives a great degree of freedom for
sound manipulation: The amplitude, frequency and phase of every partial
can be modified independantly. |
But it is often
difficult to control |
But at the same time, the fact that so
many parameters are to be generated in order to synthesize sounds, and
the fact that these parameters do not generally have a musical meaning,
makes it often difficult to control in a musical way. |
We have a solution |
Our goal is to build an instrument model
that would accept control parameters such as Pitch, Loudness, and other
timbral attibutes as inputs, and generate all the spectral parameters necessary
for the corresponding synthesis, as outputs. We want to be able to use
our model in a real-time context. |
We present results |
We used two types of models: Neural networks
and Memory-based models. We present here some very promising first results
with wind instruments (flute, saxophone), string instruments (viola, guitar)
and the singing voice. All of the neural net examples work in real-time.
|