Experimental results

Algorithm has been applied to

Singing voice

• Only 7 partials are needed to model the singing voice without artifacts.

• Room acoustics are missing $\Rightarrow$ Adapt source filter model

Piano tone

• Piano sound can be modeled with 25 partials with no audible degradation

• Decomposition of string and sound board model possible

• Model degrades when time stretched $\Rightarrow$ wrong physics?

Bass drum

• High quality for 30 partial model.

• Block transitions audible.

• Partial ramping has to be improved.

• Special Problem: Fundamental frequency is 10 times lower than for sound board signal?

  $\Rightarrow$block size trade off problem!