TTS & voice cloning

VocalBrain

In prod mlx-audioPydantic v2SQLite

What it is

TTS orchestration, 4-D model router, semantic memory (recalls the right track, zero LB import).

The architect's call

2nd consumer of the klody_memory package with zero Library Brain import — empirical proof the abstraction is right. TTS traps solved empirically (EOS never emitted on multi-sentence text → split on \n).

The code

Private repository — walkthrough over a call

Your data cannot leave the building?

That is precisely the problem I solve. A 30-minute call is enough to scope an audit.