5. Acoustic synthesis

Once the text to be produced has been analyzed, the next step is acoustic synthesis, which involves transforming the sequence of symbols obtained during linguistic analysis into a sequence of signal samples. There are two main classes of acoustic synthesis techniques: parametric and non-parametric.

In parametric synthesis, the signal is calculated using a model of the speech signal, the source-filter model: the signal passes through a "vocoder" (for "voice coder", or analysis-synthesis system).

In non-parametric synthesis, signal samples are concatenated with minimal modifications, without the need for a vocoder.