Add simple-tts example

[danemadsen]

This differs from the existing tts example by not relying on common.h and instead only using llama.h.

[danemadsen]

I have no idea why that last one is failing?

[ggerganov]

I suppose if we merge this version, we can remove the existing common-based tts example.

[ggerganov]

Note the parallel functionality is incomplete. We can either remove it to make the example simpler, or we can extend the example to support it. The latter is relatively easy to do - just store multiple sets of codes - one for each parallel sequence. And after that, generate multiple audio files - one for each set of codes.

But this can be done later to keep this PR simple - just making a note.

[danemadsen]

Yes. Also n_predict and n_parallel are constant values so they can probably be removed but I kept them in to align with the original example.

[danemadsen]

I tried to get rid of parallel processing like you suggested but it breaks the example and i cant get it to work without it. I added a not to the top of main for future devs.

Didn't want to waste too much time on it anyway as @ngxson suggests the example might not be needed at all.

[ngxson]

Tbh I think this is not needed, the tts support is very experimental for the moment, and having 2 examples for tts only adds more complexity to support for more tts models to come in the future. (Also please note that, the whole tts logic is currently built around OuteTTS. Many other tts models have different mechanism)

We should focus on faster R&D for now by relying on common, then later decide whether to move tts into a fully functional example (or even a shared library)

[danemadsen]

Though the experimental nature of TTS is a valid reason to keep using common.h, i feel that this example benefits downstream developers who are trying to build api's for this feature. I'm currently trying to implement this in dart and its very hard to do so when the only example supplied uses C++ code.