Controlling model unload for TTS

[mateuszdrab]

Is there a way to control model lifetime for TTS, for whisper WHISPER__TTL works, but I can't find a variable for TTS.
I'd like to preload a model and leave it running.

[fedirz]

For now, the WHISPER__TTL controls the TTL for Piper and Kokoro models as well. I plan on adding a separate configuration option in the future for TTS models. I'm still debating on whether this should be configured per task (i.e., you set TTL for TTS and STT) or per model (you'd set TTL for Kokoro, Piper, and Whisper separately)