Skip to content

Phonemizer

class Phonemizer

__init__

def __init__(predictor, lang_phoneme_dict)

Initializes a phonemizer with a ready predictor.

Args

  • predictor (Predictor): Predictor object carrying the trained transformer model.

  • lang_phoneme_dict (Dict[str, Dict[str, str]], optional): Word-phoneme dictionary for each language.

__call__

def __call__(text, lang, punctuation, expand_acronyms, batch_size)

Phonemizes a single text or list of texts.

Args

  • text (str): Text to phonemize as single string or list of strings.

  • lang (str): Language used for phonemization.

  • punctuation (str): Punctuation symbols by which the texts are split.

  • expand_acronyms (bool): Whether to expand an acronym, e.g. DIY -> D-I-Y.

  • batch_size (int): Batch size of model to speed up inference.

Returns
  • Union[str, List[str]]: Phonemized text as string, or list of strings, respectively.

phonemise_list

def phonemise_list(texts, lang, punctuation, expand_acronyms, batch_size)

Phonemizes a list of texts and returns tokenized texts, phonemes and word predictions with probabilities.

Args

  • texts (List[str]): List texts to phonemize.

  • lang (str): Language used for phonemization.

  • punctuation (str): Punctuation symbols by which the texts are split. (Default value = DEFAULT_PUNCTUATION)

  • expand_acronyms (bool): Whether to expand an acronym, e.g. DIY -> D-I-Y. (Default value = True)

  • batch_size (int): Batch size of model to speed up inference. (Default value = 8)

Returns
  • PhonemizerResult: Object containing original texts, phonemes, split texts, split phonemes, and predictions.

from_checkpoint

def from_checkpoint(cls, checkpoint_path, device, lang_phoneme_dict)

Initializes a Phonemizer object from a model checkpoint (.pt file).

Args

  • checkpoint_path (str): Path to the .pt checkpoint file.

  • device (str): Device to send the model to ('cpu' or 'cuda'). (Default value = 'cpu')

  • lang_phoneme_dict (Dict[str, Dict[str, str]], optional): Word-phoneme dictionary for each language.

Returns
  • Phonemizer: Phonemizer object carrying the loaded model and, optionally, a phoneme dictionary.