i am currently implementing the bandwidth enhanced sinusoidal model for an additive synthesizer, this model allow to accurately produce noisy sounds by adding a noise component to each sinusoids.
But i have some troubles understanding the technical side of that model, to me, it seem that a small amount of filtered white noise is added to each partials so that each bands are widened.
So, in my implementation, white noise samples ([-1, 1] random values, gaussian distribution) is taken from a white noise wavetable, a low-pass filter (Chebychev order 3, cutoff 500, ripple -1., i also tried a simple moving average filter) is then applied on the resulting noise samples, the filtered noise value is then used as a phase step multiplier for the sine oscillators, this actually spread the frequency of the oscillator around as pictured below.
However, the audio result is a kind of background noise and does not enhance the noise part of the sound at all, even with an envelope applied to the noise multiplier.
What is wrong/different with this implementation compared to the model in the article linked above ?