GitHub - allo-media/text2num: Parse and convert numbers written in French, English, Spanish, Portuguese, German, Dutch and Italian into their digit representation.

A fast and robust text2num converter library a library for recognizing, parsing and transcribing into digits (base 10) numbers expressed in natural language. It works on strings as well as on custom token lists.

No IA involved: it is low on resources (and energy!) consumption and the latency is very small.

text2num is a python package that provides functions and parser classes for:

Parsing numbers expressed as natural language words and converting them to integer values.
Detection of ordinal, cardinal and decimal numbers in a stream of natural language words and get their decimal digit representations.

Supported natural languages (in alphabetical order):

Dutch;
English;
French;
German;
Italian;
Portuguese (Brazilian and European);
Spanish.

Versions 3.X vs 2.X

This new generation of text2num relies on a new and improved algorithm implemented in Rust whereas the 2.X branch is in pure python and uses a less capable algorithm and has now been retired.

You don't need Rust to install and run text2num! as we provide precompiled wheels.

Backward incompatible changes:

dropped support for signed numbers — the feature was broken anyway;
parsing mode is relaxed by default (i.e. "greedy") — you can use punctuation (e.g. commas) to separate groups in text, or voice pauses if processing Speech-to-Text token streams;
the threshold optional parameter to alpha2digits now applies to both ordinals and cardinals. As a consequence the signature of alpha2digits has changed.
the Russian and Catalan languages have not been ported yet.

Installation

We provide pre-compiled wheels for Linux, MacOS and Windows:

	py 3.8	py 3.9	py 3.10	py 3.11	py 3.12	py 3.13
Linux aarch64	✅	✅	✅	✅	✅	✅
Linux armv7l	✅	✅	✅	✅	✅	✅
Linux ppc64le	✅	✅	✅	✅	✅	✅
Linux s390x	✅	✅	✅	✅	✅	✅
Linux x86_64	✅	✅	✅	✅	✅	✅
Linux i686	✅	✅	✅	✅	✅	✅
win32	✅	✅	✅	✅	✅	✅
win amd64	✅	✅	✅	✅	✅	✅
macos 11	❌	❌	❌	✅	✅	✅

So all you need to do is this:

pip install text2num
# or, if using an `uv` project:
uv add text2num

Quickstart by example

Not every supported language is covered in these examples, but it gives you an idea.

Parse and convert

French examples:

    >>> from text_to_num import text2num
    >>> text2num('quatre-vingt-quinze', "fr")
    95

    >>> text2num('nonante-cinq', "fr")
    95

    >>> text2num('mille neuf cent quatre-vingt dix-neuf', "fr")
    1999

    >>> text2num('dix-neuf cent quatre-vingt dix-neuf', "fr")
    1999

    >>> text2num("cinquante et un million cinq cent soixante dix-huit mille trois cent deux", "fr")
    51578302

    >>> text2num('mille mille deux cents', "fr")
    Traceback (most recent call last):
        ...
    ValueError: invalid literal for text2num: 'mille mille deux cents'

English examples:

    >>> from text_to_num import text2num

    >>> text2num("fifty-one million five hundred seventy-eight thousand three hundred two", "en")
    51578302

    >>> text2num("eighty-one", "en")
    81

Spanish examples:

    >>> from text_to_num import text2num
    >>> text2num("ochenta y uno", "es")
    81

    >>> text2num("nueve mil novecientos noventa y nueve", "es")
    9999

    >>> text2num("cincuenta y tres millones doscientos cuarenta y tres mil setecientos veinticuatro", "es")
    53243724

Portuguese examples:

    >>> from text_to_num import text2num
    >>> text2num("trinta e dois", "pt")
    32

    >>> text2num("mil novecentos e seis", "pt")
    1906

    >>> text2num("vinte e quatro milhões duzentos mil quarenta e sete", "pt")
    24200047

German examples:

    >>> from text_to_num import text2num

    >>> text2num("einundfünfzigmillionenfünfhundertachtundsiebzigtausenddreihundertzwei", "de")
    51578302

    >>> text2num("ein und achtzig", "de")
    81

Find and transcribe

Any numbers, even ordinals.

French:

    >>> from text_to_num import alpha2digit
    >>> sentence = (
    ...     "Huit cent quarante-deux pommes, vingt-cinq chiens, mille trois chevaux, "
    ...     "douze mille six cent quatre-vingt-dix-huit clous.\n"
    ...     "Quatre-vingt-quinze vaut nonante-cinq. On tolère l'absence de tirets avant les unités : "
    ...     "soixante seize vaut septante six.\n"
    ...     "Nombres en série : douze, quinze, zéro zéro quatre, vingt, cinquante-deux, cent trois, cinquante deux, "
    ...     "trente et un.\n"
    ...     "Ordinaux: cinquième troisième vingt et unième centième mille deux cent trentième.\n"
    ...     "Décimaux: douze virgule quatre-vingt dix-neuf, cent vingt virgule zéro cinq ; "
    ...     "mais soixante zéro deux."
    ... )
    >>> print(alpha2digit(sentence, "fr"))
    842 pommes, 25 chiens, 1003 chevaux, 12698 clous.
    95 vaut 95. On tolère l'absence de tirets avant les unités : 76 vaut 76.
    Nombres en série : 12, 15, 004, 20, 52, 103, 52, 31.
    Ordinaux: 5ème 3ème 21ème 100ème 1230ème.
    Décimaux: 12,99, 120,05 ; mais 60 02.

    >>> sentence = "Cinquième premier deuxième troisième vingt et unième centième mille deux cent trentième."
    >>> print(alpha2digit(sentence, "fr"))
    5ème 1er 2ème 3ème 21ème 100ème 1230ème.

English:

    >>> from text_to_num import alpha2digit

    >>> text = "On May twenty-third, I bought twenty-five cows, twelve chickens and one hundred twenty five point four zero kg of potatoes."
    >>> alpha2digit(text, "en")
    'On May 23rd, I bought 25 cows, 12 chickens and 125.40 kg of potatoes.'

    >>> alpha2digit("I finished the race in the twelfth position!", "en")
    'I finished the race in the 12th position!'

Spanish:

    >>> from text_to_num import alpha2digit

    >>> text = "Compramos veinticinco vacas, doce gallinas y ciento veinticinco coma cuarenta kg de patatas."
    >>> alpha2digit(text, "es")
    'Compramos 25 vacas, 12 gallinas y 125,40 kg de patatas.'

    >>> text = "Compramos veinticinco vacas, doce gallinas y ciento veinticinco punto cuarenta kg de patatas."
    >>> alpha2digit(text, "es")
    'Compramos 25 vacas, 12 gallinas y 125.40 kg de patatas.'

    >>> text = "Ella ha quedado tercera"
    >>> alpha2digit(text, "es", threshold=0)
    'Ella ha quedado 3ª'

Portuguese:

    >>> from text_to_num import alpha2digit

    >>> text = "Comprámos vinte e cinco vacas, doze galinhas e cento e vinte e cinco vírgula quarenta kg de batatas."
    >>> alpha2digit(text, "pt")
    'Comprámos 25 vacas, 12 galinhas e 125,40 kg de batatas.'

    >>> text = "Ordinais: quinto, terceiro, vigésima, vigésimo primeiro, centésimo quarto"
    >>> alpha2digit(text, "pt")
    'Ordinais: 5º, 3º, 20ª, 21º, 104º'

German:

    >>> from text_to_num import alpha2digit

    >>> text = "Ich habe fünfundzwanzig Kühe, zwölf Hühner und einhundertfünfundzwanzig kg Kartoffeln gekauft."
    >>> alpha2digit(text, "de")
    'Ich habe 25 Kühe, 12 Hühner und 125 kg Kartoffeln gekauft.'

    >>> text = "Die Telefonnummer lautet dreiunddreißig neun sechzig null sechs zwölf einundzwanzig."
    >>> alpha2digit(text, "de")
    'Die Telefonnummer lautet 33 9 60 06 12 21.'

    >>> text = "Der zweiundzwanzigste Januar zweitausendzweiundzwanzig."
    >>> alpha2digit(text, "de")
    'Der 22. Januar 2022.'

    >>> text = "Es ist ein Buch mit dreitausend Seiten aber nicht das erste."
    >>> alpha2digit(text, "de", threshold=0)
    'Es ist 1 Buch mit 3000 Seiten aber nicht das 1..'

    >>> text = "Pi ist drei Komma eins vier und so weiter, aber nicht drei Komma vierzehn :-p"
    >>> alpha2digit(text, "de", threshold=0)
    'Pi ist 3,14 und so weiter, aber nicht 3 Komma 14 :-p'

Working with tokens

Imagine that we have an ASR application that returns a transcript as a list of tokens (text, start timestamp, end timestamp) where the timestamps are integers representing milliseconds relative to the beginning of the speech.

from text_to_num import (Token, find_numbers)


class DecodedWord(Token):
    def __init__(self, text, start, end):
        self._text = text
        self.start = start
        self.end = end

    def text(self):
        return self._text

    def nt_separated(self, previous):
        # we consider a voice gap of more that 100 ms as significant
        return self.start - previous.end > 100


# Let's simulate ASR output

stream = [
    DecodedWord("We", 0, 100),
    DecodedWord("have", 100, 200),
    DecodedWord("respectively", 200, 400),
    DecodedWord("twenty", 400, 500),
    DecodedWord("nine", 610, 700),
    DecodedWord("and", 700, 800),
    DecodedWord("thirty", 800, 900),
    DecodedWord("four", 950, 1000),
    DecodedWord("dollars", 1010, 1410)
]

occurences = find_numbers(stream, "en")

for num in occurences:
    print(f"found number {num.text} ({num.value}) at range [{num.start}, {num.end}] in the stream")

When executed, that code snippet prints::

found number 20 (20.0) at range [3, 4] in the stream
found number 9 (9.0) at range [4, 5] in the stream
found number 34 (34.0) at range [6, 8] in the stream

Read the complete documentation on ReadTheDocs.

Contribute

Join us on https://github.com/allo-media/text2num