add delta related functionality

replace torchaudio with soundfile for audio loading in run_inference

remove unused quantization configuration from model loading

update model architecture to the latest

enhance run_inference to support token masking and adjust logits handling

add logger + replace threading with multiprocessing

add duration scores functionality

do tokenization manually

add is_split_into_words argument to correctly handle the list of phonemes

pass phonemes list instead of transcript + csimplify score calculation logic

change token_lengths calculation

add phoneme validation + refactoring + constant in separate file

validate audio + refactoring