num2words2 1.0.18

Performance

Because the conversion engine is compiled Rust, num2words2 is several times faster than the original pure-Python num2words. Benchmarked against num2words 0.5.14, Apple M-series, nanoseconds per call:

The speedup is largest where the original does heavy Python-level parsing (string inputs) or deep recursion (French, German); it never comes at the cost of accuracy — output is byte-for-byte identical to the original across a frozen ~150,000-case corpus of cardinals, ordinals, currency, years, fractions, floats and strings.

Installation

Install from PyPI — the wheels bundle the Python binder and the compiled Rust extension, built per platform:

pip install num2words2

Wheels for every version are also attached to the GitHub Releases page; the compiled extension is distributed only through those release wheels (it is not committed to the repository).

To build from source you need a stable Rust toolchain and maturin:

maturin build --release

Development Setup

The project uses pre-commit hooks to ensure code quality. To set up your development environment:

# Install pre-commit
pip install pre-commit

# Install the git hook scripts
pre-commit install

# Run hooks on all files (optional, useful for initial setup)
pre-commit run --all-files

This will automatically format and lint your code before each commit using:

• autopep8 - PEP 8 formatting

• autoflake - removes unused imports and variables

• isort - sorts imports

• flake8 - style and quality checks

• trailing-whitespace removal

• end-of-file fixing

Testing

The library uses pytest for testing. First, install the development dependencies:

make dev-install

Then, you can run the test suite using several methods:

• Run basic tests: This runs tests with your current Python environment.

  make test

• Run with Tox: This runs tests against all supported Python versions, which is the standard for CI.

  tox

python tests/scripts/generate_llm_tests.py --languages fr,es --samples 10

Usage

Command line:

$ num2words2 10001
ten thousand and one
$ num2words2 24,120.10
twenty-four thousand, one hundred and twenty point one
$ num2words2 24,120.10 -l es
veinticuatro mil ciento veinte punto uno
$ num2words2 2.14 -l es --to currency
dos euros con catorce céntimos

In code there’s only one function to use:

>>> from num2words2 import num2words
>>> num2words(42)
forty-two
>>> num2words(42, to='ordinal')
forty-second
>>> num2words(42, lang='fr')
quarante-deux

Besides the numerical argument, there are two main optional arguments, to: and lang:

to: The converter to use. Supported values are:

• cardinal (default)

• ordinal

• ordinal_num

• year

• currency

lang: The language in which to convert the number. Supported values are:

• en (English, default)

• am (Amharic)

• ar (Arabic)

• az (Azerbaijani)

• be (Belarusian)

• bn (Bangladeshi)

• ca (Catalan)

• ce (Chechen)

• cs (Czech)

• cy (Welsh)

• da (Danish)

• de (German)

• en_GB (English - Great Britain)

• en_IN (English - India)

• en_NG (English - Nigeria)

• es (Spanish)

• es_CO (Spanish - Colombia)

• es_CR (Spanish - Costa Rica)

• es_GT (Spanish - Guatemala)

• es_VE (Spanish - Venezuela)

• eu (EURO)

• fa (Farsi)

• fi (Finnish)

• fr (French)

• fr_BE (French - Belgium)

• fr_CH (French - Switzerland)

• fr_DZ (French - Algeria)

• he (Hebrew)

• hi (Hindi)

• hu (Hungarian)

• hy (Armenian)

• id (Indonesian)

• is (Icelandic)

• it (Italian)

• ja (Japanese)

• kn (Kannada)

• ko (Korean)

• kz (Kazakh)

• mn (Mongolian)

• lt (Lithuanian)

• lv (Latvian)

• nl (Dutch)

• no (Norwegian)

• pl (Polish)

• pt (Portuguese)

• pt_BR (Portuguese - Brazilian)

• ro (Romanian)

• ru (Russian)

• sl (Slovene)

• sk (Slovak)

• sr (Serbian)

• sv (Swedish)

• te (Telugu)

• tet (Tetum)

• tg (Tajik)

• tr (Turkish)

• th (Thai)

• uk (Ukrainian)

• vi (Vietnamese)

• zh (Chinese - Traditional)

• zh_CN (Chinese - Simplified / Mainland China)

• zh_TW (Chinese - Traditional / Taiwan)

• zh_HK (Chinese - Traditional / Hong Kong)

You can supply values like fr_FR; if the country doesn’t exist but the language does, the code will fall back to the base language (i.e. fr). If you supply an unsupported language, NotImplementedError is raised. Therefore, if you want to call num2words with a fallback, you can do:

try:
    return num2words(42, lang=mylang)
except NotImplementedError:
    return num2words(42, lang='en')

Additionally, some converters and languages support other optional arguments that are needed to make the converter useful in practice.

Wiki

For the full documentation, including installation, API details, CLI usage, supported locales, sentence conversion, currency handling, and migration guidance, please check the Wiki. Feel free to propose wiki enhancements.

History

num2words is based on an old library, pynum2word, created by Taro Ogawa in 2003. Unfortunately, the library stopped being maintained and the author can’t be reached. There was another developer, Marius Grigaitis, who in 2011 added Lithuanian support, but didn’t take over maintenance of the project.

Virgil Dupras from Savoir-faire Linux based himself on Marius Grigaitis’ improvements and re-published pynum2word as num2words.

Relationship to num2words

num2words2 is not a divergent fork. It is a Rust port of the num2words conversion engine with a Python binder, kept output-compatible with the original.

The original project’s issues and pull requests are actively monitored. Valid, relevant fixes and features are reviewed and then ported into this project’s Rust core, with byte-parity re-verified against the frozen corpus. num2words2 therefore tracks upstream improvements while delivering them natively — and several times faster.

Jean-Louis Queguiner