In speech signal processing, the source filter model is a mathematical model that represents the [[speech signal]] by a combination of a sound source with a linear acoustic filter.
The sound source for voiced sounds needs to model the [[glottal airflow velocity]]. For unvoiced sounds, the source needs to model the noise produced by the turbulent airflow at a vocal constriction.
The acoustic filter aims to model the way the vocal tract shapes the spectral envelope of speech sounds. It is often modeled by a [[discrete-time LTI system]].
A simplified model can be represented as:
![[source-filter-model.png]]
[[Huang 2001]]
In this model, the source can be an impulse train to simulate the [[voiced phonation]] or white noise for the [[unvoiced phonation]]. More complex models use a weighted sum of periodic and non-periodic signals for the source signal.