Create modified model

[avinoam134]

Hey all! im quite new to deep learning models (for both applicatative and development aspects) and not the most experienced developer overall anyway.
However im keen to learn by creating a big project, involving trying some cool things with this incredible library - one of them includes modifying a model that is already supported.
for example - say i create a new kokoro_tts.onnx that returns another parameter, in addition to the same audio that is already generated (keeping the pre-trained weights still compatible and used the same way).
Where do i need to look for things to modify so that this when my new model is ran, the extra output value is propagated back to whoever uses the TTS API?