g2p_en Documentation

The conversion process follows these steps:

1. Normalization: Spells out Arabic numbers and currency symbols (e.g., $200 becomes two hundred dollars).
2. Disambiguation: Attempts to retrieve correct pronunciations for heteronyms based on their Part-of-Speech (POS).
3. Dictionary Lookup: Uses the CMU Pronouncing Dictionary for standard words.
4. Neural Prediction: For Out-of-Vocabulary (OOV) words, it predicts pronunciations using a neural net model (inference is performed using NumPy).

You can install the g2p_en package using pip. Note that the nltk package dependencies will be automatically downloaded during your first run.

pip install g2p_en

To convert English text (graphemes) to phonemes, import the G2p class from g2p_en. You can instantiate the class and call the instance directly with a string. The module handles:

• Spelling out Arabic numbers and currency symbols.
• Disambiguating homographs using Part-of-Speech (POS) tagging.
• Looking up words in the CMU Pronouncing Dictionary.
• Predicting pronunciations for Out-of-Vocabulary (OOV) words using a neural network model.

The output is a list of phonemes and spaces.

from g2p_en import G2p

texts = [
    "I have $250 in my pocket.",
    "popular pets, e.g. cats and dogs",
    "I refuse to collect the refuse around here.",
    "I'm an activationist."
]

g2p = G2p()
for text in texts:
    out = g2p(text)
    print(out)

The following dependencies are required for g2p_en:

• numpy >= 1.13.1
• nltk >= 3.2.4 (requires averaged_perceptron_tagger and cmudict NLTK data)
• inflect >= 0.3.1
• Distance >= 0.1.3