limitation of source-filter model

[Ruiduan L]

It is quite reasonable that both vocal folds and friction could be the sources of sounds and in the video, it has been mentioned that some languages have neither of these things happened and some have both of them. So what is the relationship between this phenomenon and the limitation of source-filter model?

[Simon]

The simplest version of the source-filter model only uses one source at a time (either periodic, or non-periodic). This cannot model voiced fricatives, so we need to upgrade the source-filter model to have mixed excitation.

This is still not a great model though, because the voiced an unvoiced sources will be shaped by the same filter, whereas in the vocal tract the two sources are often at different physical locations and so have a different amount of vocal tract between the source and the lips.

Sounds like clicks and even plosive bursts are not well-modelled by a simple source-filter model.

But, in the end, we need to stress that the source-filter model is a model of the speech signal (that’s all we need) and not a faithful model of the physics of speech production (which would be interesting, but not essential for our purposes).

[Author]

Posts