english2kana

english2kana is a Python library designed to convert English words and romanized Japanese text into their Katakana representations. It leverages a sequence-to-sequence model with an attention mechanism, trained on data derived from English company names and their corresponding Katakana forms found on the Japan Corporate Number Publication Site.

Key Features:

• Converts English words to Katakana.
• Employs a seq2seq model architecture with attention.
• Achieves high accuracy with the given dataset.

Model Performance:

• Test Loss: 0.1440
• Test Accuracy: 0.9552

Requirements

Python 3.11 or higher

Installation

pip install english2kana

Usage

from english2kana import english2kana

# Initialize the translator
e2k = english2kana()
# Load the pretrained model
e2k.load_model()

# Translate an English word into Katakana
output = e2k.translate('simple')
print(output)  # シンプル

Data

The training data is sourced from the Japan Corporate Number Publication Site. The dataset includes a wide array of English corporate names along with their correct Katakana representations, ensuring the model is exposed to various letter combinations and phonetic patterns.

Contributing

Contributions are welcome! Feel free to open issues or submit pull requests to help improve this library.

License

The MIT License (MIT)