This package is a fork of fiato by cfahlgren1 (https://huggingface.co/cfahlgren1/fiato), used under the Apache License 2.0. fiato is a fork of tiny-voca by Muhammed Nazeem (https://github.com/n4ze3m/tiny-voca, commit 62e2ade), used under the Apache License 2.0. The original license is in LICENSE.tiny-voca. Changes in fiato: - Expression tokens (, , ...) in the text front end, the aligner and the annotation checks. - Two new voices, F1 and M1. - Feature preparation runs on several processes. - Support for tiny-voca's earlier byte-token feature files is removed. Changes in this fork (pitput): - Hebrew front end: Phonikud phoneme labels instead of eSpeak/dizione. - Eight delivery styles in place of the voices; whispered styles are not forced voiced. - Eight expression tags. - Features stored as 16-bit waveform and float16 mel; training can stream them from disk with background prefetching. - Inference reads Hebrew text through Phonikud.