English Grapheme To Phoneme Conversion

2.1.0 · maintenance · verified Mon Apr 13

g2p-en is a Python module designed for converting English graphemes (spelling) to phonemes (pronunciation). It is essential for tasks like speech synthesis. The library uses a combination of dictionary lookups, part-of-speech tagging for homograph disambiguation, and a neural network (using NumPy for inference as of v2.0) for out-of-vocabulary words. The current version is 2.1.0, released in late 2019, and its release cadence appears to be infrequent.

Warnings

breaking Version 2.0 removed TensorFlow as a dependency, replacing it with NumPy for neural network inference. Users upgrading from pre-2.0 versions will no longer require TensorFlow.
Fix: Ensure your environment is set up for NumPy. If you had TensorFlow installed specifically for g2p-en, it's no longer needed.
gotcha The library requires downloading specific NLTK data files ('averaged_perceptron_tagger' and 'cmudict') after installation. This step is crucial for the library's functionality and is not performed automatically by `pip install`.
Fix: Run `python -m nltk.downloader "averaged_perceptron_tagger" "cmudict"` after installing the package.
gotcha While the library attempts to disambiguate homographs (words spelled the same but pronounced differently, like 'refuse' as a verb vs. noun) using part-of-speech tagging, perfect contextual disambiguation is not always guaranteed for all cases.
Fix: Be aware of potential ambiguities for homographs in critical applications; manual review or external context might be necessary.
gotcha For words not present in its internal dictionaries (Out-Of-Vocabulary words), g2p-en uses a neural network to predict pronunciations. While it makes a 'best guess', accuracy may vary for highly novel, specialized, or irregularly spelled terms.
Fix: Test with a representative corpus of OOV words if accuracy is critical; consider adding custom dictionary entries if specific OOV words are consistently mispronounced.
gotcha The library performs internal text preprocessing, including spelling out numbers ($250 -> two hundred fifty dollars), expanding common abbreviations (e.g. -> for example), and normalizing contractions (I'm -> I am). This can alter the input text before G2P conversion.
Fix: Be aware that input text is transformed. If fine-grained control over text normalization is needed, consider pre-processing your text externally before passing it to `g2p-en`.

Install

pip install g2p-en Install core library
python -m nltk.downloader "averaged_perceptron_tagger" "cmudict" Download NLTK data (required post-install)

Imports

G2p
```
from g2p_en import G2p
```

Quickstart

Initializes the G2p converter and processes a list of English sentences, demonstrating handling of numbers, abbreviations, homographs, and out-of-vocabulary words.

from g2p_en import G2p

texts = [
    "I have $250 in my pocket.", # number -> spell-out
    "popular pets, e.g. cats and dogs", # e.g. -> for example
    "I refuse to collect the refuse around here.", # homograph
    "I'm an activationist." # newly coined word
]

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

view raw JSON →