import ModelInterfaces import torch import numpy as np import epitran import eng_to_ipa class EpitranPhonemConverter(ModelInterfaces.ITextToPhonemModel): word_locations_in_samples = None audio_transcript = None def __init__(self, epitran_model) -> None: super().__init__() self.epitran_model = epitran_model def convertToPhonem(self, sentence: str) -> str: phonem_representation = self.epitran_model.transliterate(sentence) return phonem_representation class EngPhonemConverter(ModelInterfaces.ITextToPhonemModel): def __init__(self,) -> None: super().__init__() def convertToPhonem(self, sentence: str) -> str: phonem_representation = eng_to_ipa.convert(sentence) phonem_representation = phonem_representation.replace('*','') return phonem_representation