ComputerysProfanityFilter 1.1.0

ComputerysProfanityFilter

A .NET Standard 2.1 (C# version 9 for unity) profanity filter that censors known terms while recognizing common obfuscations such as character substitutions, punctuation, inserted whitespace, repeated letters, and leetspeak. The default term list is reasonable and covers swears, hate speech, and self-harm.

This library is currently used in STRAFTAT (as of the 1.4.9 update) to filter text chat, player names, lobby names, etc, albeit with a slightly cut down word list (swears removed).

Installation

NuGet

Install the package with the .NET CLI:

dotnet add package ComputerysProfanityFilter

Or add ComputerysProfanityFilter through the NuGet package manager in your IDE.

Unity Package Manager

In Unity, open Window > Package Manager, select +, then choose Add package from git URL..., then enter:

https://github.com/C0mputery/ComputerysProfanityFilter.git?path=/ComputerysProfanityFilter#upm-1.1.0

The #upm-1.1.0 suffix pins the package to version 1.1.0; change it to a newer upm-x.y.z tag to update. The UPM package requires Unity 2021.3 or newer.

Basic usage

Create one ProfanityList and reuse it when filtering messages. After construction, the instance can be used concurrently from multiple threads. Censor returns the original string unchanged when it finds no match.

using ComputerysProfanityFilter;

ProfanityList profanityFilter = new ProfanityList();
string censored = profanityFilter.Censor("Please don't be an a$$hole.");
// censored: "Please don't be an #######."

Use the optional second argument to choose the replacement character:

string censored = profanityFilter.Censor("Please don't be an a$$hole.", '*');
// censored: "Please don't be an *******."

Matching flexable, case-insensitive, it'll normalizes many substitutions (1337 speak, common enlgish rules), ignores selected punctuation within words, and collapses consecutive repeated letters. Entire matches, including their intervening punctuation, are replaced with the censor character.

Matching flexable, case-insensitive, it'll normalizes many substitutions (1337 speak, common enlgish rules). Punctuation and whitespace may be inserted within terms without preventing a match. Consecutive repeated letters are collapsed, so variants such as foooool can match fool. Matchs are as terms rather than arbitrary substrings, avoiding the Scunthorpe problem, cunt does not match inside Scunthorpe. Recognized punctuation can also act as a boundary between terms. Entire matches, including intervening whitespace and punctuation, are replaced with the censor character.

Detecting profanity

Use HasProfanity when you only need a yes/no result:

bool containsProfanity = filter.HasProfanity("That is a$$hole behavior.");

Use DetectAllProfanities when you need the locations and configured terms that matched:

var matches = filter.DetectAllProfanities("Sh!t happens.");

ProfanityMatch match = matches[0];
// match.Start: 0
// match.End: 4 (exclusive)
// match.Term: "shit"

Start is the zero-based inclusive index in the original input. End is the zero-based exclusive index. A match's Term is the configured term, even when the input uses an expanded or obfuscated form. Matches are returned in source order. Allow terms are excluded from all three operations.

Custom term list

Pass your own terms and normalization rules to the constructor. By default, supplied terms are expanded into several common English forms, so use base terms where appropriate. Set expandTermForms: false to match only the supplied forms after normalization.

using System.Collections.Generic;
using ComputerysProfanityFilter;

ProfanityList filter = new ProfanityList(
    terms: new[] { "spoiler", "example phrase" },
    allowTerms: new[] { "allowed word" },
    expandTermForms: false,
    expectedCharacters: DefaultProfanityList.ExpectedCharacters,
    joinerCharacters: DefaultProfanityList.JoinerCharacters,
    boundaryCharacters: new[] { '.', '-', '_' },
    characterMap: new Dictionary<char, string> { ['4'] = "a", ['@'] = "a" },
    sequenceMap: new Dictionary<string, string>()
);

string censored = filter.Censor("S-p-o-i-l-e-r ahead");
// censored: "############# ahead"

Default configuration

DefaultProfanityList exposes the built-in settings as read-only collections for reuse in custom filters: Terms, AlwaysCensorTerms, AllowTerms, ExpectedCharacters, JoinerCharacters, BoundaryCharacters, CharacterMap, and SequenceMap. Allow terms are compared to the original matched text using lowercase-invariant normalization only.

Allocation behavior

Censor is zero allocate on the heap when no term matches, it'll return the same string instance that was passed in. Only when a match is found will it rent a buffer from ArrayPool<char>.Shared and returns a string that is the same length as the input. The rented buffers can be reused across calls after the pool is warm, although rent may allocate when the pool has no suitable buffer available (and there are some other caveats with ArrayPool, but) Under typical use conditions, it will only allocate a string of the same size input.

Comparisons

Benchmarks

BenchmarkDotNet 0.15.8 results with .NET 10.0.10 on an AMD Ryzen 7 5800X (Windows 11). The suite compares this library with Profanity.Detector 0.1.8, BogaNet.BadWordFilter, DotnetBadWordDetector, Censored, and mk.profanity, using equivalent vocabularies where possible. Inputs are exactly 100, 1,000, 10,000, or 100,000 characters, made by repeating a fixed 10-message chat corpus containing ordinary, profane, and obfuscated text.

Benchmark run date: August 21, 2026.

Censoring

Results are grouped by vocabulary so each comparison uses the same term list. Times are mean time per Censor operation; allocation is managed allocation per operation.

Expanded default vocabulary (2,573 terms)

Profanity.Detector raw vocabulary (1,617 terms)

CPU-stress input

This benchmark uses a repeated аss… input. CPU-stress numbers re-run August 21, 2026.

Construction

Construction measurements create one filter instance. This system doesn't have free construction, so be a little bit wary of that. Make sure that you cache your instance after it's created.

License

Copyright 2026 Christopher Rohland. This project is licensed under the Apache License 2.0. See the LICENSE file for the full license text.