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.

usingComputerysProfanityFilter;ProfanityListprofanityFilter=newProfanityList();stringcensored=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:

stringcensored=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:

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

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

varmatches=filter.DetectAllProfanities("Sh!t happens.");ProfanityMatchmatch=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.

usingSystem.Collections.Generic;usingComputerysProfanityFilter;ProfanityListfilter=newProfanityList(terms:new[]{"spoiler","example phrase"},allowTerms:new[]{"allowed word"},expandTermForms:false,expectedCharacters:DefaultProfanityList.ExpectedCharacters,joinerCharacters:DefaultProfanityList.JoinerCharacters,boundaryCharacters:new[]{'.','-','_'},characterMap:newDictionary<char,string>{['4']="a",['@']="a"},sequenceMap:newDictionary<string,string>());stringcensored=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

Feature	ComputerysProfanityFilter	Profanity.Detector	BogaNet.BadWordFilter	DotnetBadWordDetector	Censored	mk.profanity
Distribution	NuGet + UPM (see above)	NuGet	NuGet	NuGet	NuGet	NuGet
Matching approach	Deterministic normalized term matching	Dictionary/regex scanning + Scunthorpe heuristic	Regex corpus + optional normalization; optional simple matching	ML.NET binary classifiers	Escaped wildcard patterns compiled to regex	Tokenized exact list + optional Levenshtein
Case-insensitive matching	✓	✓	✓	◐ It's ML	✓	◐
Word-boundary	✓	◐	◐	◐ It's ML	✓	◐
Leetspeak normalization	✓	✗	✓	◐ It's ML	✗	✗
Inserted punctuation handling	✓	◐	◐	✓	◐	✗
Inserted spaces handling	✓	✗	◐	◐ It's ML	◐	✗
Repeated-letter handling	✓	✗	✗	◐ It's ML	✗	◐
Multi-word phrase support	✓	✓	✓	◐	✓	◐
Match positions	✓	✗	✗	✗	✗	✓
Custom list	✓	✓	✓	✗	✓	✓
Allow list	✓	✓	✗	✗	✗	✓
Configurable censor character	✓	✓	✓	✓	✗	✓
Arbitrary custom-language terms/rules	✓	✓	✓	✗ It's ML	✓	✓
Concurrency/thread safety	✓ Explicitly documented	Mutable, not documented	Mutable, not documented	✗ Uses PredictionEngine	Mutable, not documented	Mutable, not documented
External dependencies	None	None	BogaNet.* infrastructure packages	Microsoft.ML 1.7.0	None	None
Trimming/AOT compatibility	✓	✓	◐	✗	✓	✓
Bundled profanity data	✓ 145 terms (2,573 with variations)	✓ 1,617 terms, the list is weird, read it, very much quantity over quality)	✓ ~4,600 entries (claims 5,000+ regexes)	✓ 4 ML Models	✗	✓ 2,915 words
Bundled languages	English	English	25	4	0	English
Fuzzy matching	✗	✗	✗	◐ It's ML	◐	✓
Confidence	✗	✗	✗	✓	✗	✗
Mutable	✗	✓	✓	✗	✓	◐
Patterns	✗	✗	✓ Regex	✗	✓ Wildcards	✗
URL/email moderation	✗	✗	✓	✗	✗	✗
Other symbol/content filters	✗	✗	✓	✗	✗	✗
Async rule loading	✗	✗	✓	✗	✗	✗
Target framework	.NET Standard 2.1 and .NET 10	.NET Standard 2.0	.NET 8	.NET 8	.NET Standard 1.0	.NET Standard 2.1
License	Apache-2.0	MIT	MIT	Apache-2.0	MIT	MIT

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, 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)

Mean Time	100 chars	1,000 chars	10,000 chars	100,000 chars
ComputerysProfanityFilter	2.468 us	29.422 us	302.077 us	2.981 ms
Profanity.Detector	109.157 us	1.165 ms	113.736 ms	1.943 s
BogaNet.BadWordFilter	151.999 us	1.464 ms	14.817 ms	166.816 ms
DotnetBadWordDetector	12.619 us	118.059 us	1.175 ms	12.425 ms
Censored	8.154 ms	8.146 ms	9.929 ms	25.445 ms
mk.profanity	51.646 us	391.885 us	3.838 ms	39.767 ms

Managed Allocation	100 chars	1,000 chars	10,000 chars	100,000 chars
ComputerysProfanityFilter	-	2 KB	20 KB	195 KB
Profanity.Detector	8 KB	268 KB	68.00 MB	2.40 GB
BogaNet.BadWordFilter	520 B	29 KB	2.04 MB	205.42 MB
DotnetBadWordDetector	3 KB	24 KB	222 KB	2.15 MB
Censored	14.09 MB	14.17 MB	14.29 MB	15.52 MB
mk.profanity	4 KB	19 KB	151 KB	1.42 MB

Profanity.Detector raw vocabulary (1,617 terms)

Mean time	100 chars	1,000 chars	10,000 chars	100,000 chars
ComputerysProfanityFilter	2.903 us	33.929 us	336.551 us	3.429 ms
Profanity.Detector	72.138 us	751.393 us	73.684 ms	1.379 s
BogaNet.BadWordFilter	1.188 ms	9.846 ms	99.245 ms	1.041 s
DotnetBadWordDetector	13.296 us	117.788 us	1.173 ms	12.609 ms
Censored	2.872 ms	3.163 ms	4.114 ms	12.828 ms
mk.profanity	41.420 us	340.211 us	3.452 ms	34.629 ms

Managed Allocation	100 chars	1,000 chars	10,000 chars	100,000 chars
ComputerysProfanityFilter	-	2 KB	20 KB	195 KB
Profanity.Detector	5 KB	290 KB	49.33 MB	2.11 GB
BogaNet.BadWordFilter	520 B	49 KB	2.98 MB	294.18 MB
DotnetBadWordDetector	3 KB	24 KB	222 KB	2.15 MB
Censored	5.65 MB	5.70 MB	5.92 MB	8.13 MB
mk.profanity	4 KB	20 KB	152 KB	1.42 MB

CPU-stress input

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

Mean Time	100 chars	1,000 chars	10,000 chars	100,000 chars
ComputerysProfanityFilter	6.768 us	73.650 us	761.768 us	7.955 ms

Managed Allocation	100 chars	1,000 chars	10,000 chars	100,000 chars
ComputerysProfanityFilter	224 B	2 KB	20 KB	195 KB

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.

Engine and vocabulary	Mean	Allocated
ComputerysProfanityFilter / expanded default (2,573 terms)	4.616 ms	5.39 MB
Profanity.Detector / same expanded vocabulary (2,573 terms)	5.490 us	32.98 KB
ComputerysProfanityFilter / Profanity.Detector raw vocabulary (1,617 terms)	1.364 ms	2.90 MB
Profanity.Detector / raw default vocabulary (1,617 terms)	4.399 us	25.58 KB
BogaNet.BadWordFilter / English sources	2.225 ms	1.64 MB
DotnetBadWordDetector / default model	29.899 ms	2.91 MB
Censored / equivalent expanded vocabulary	770.0 ns	20.18 KB
mk.profanity / equivalent expanded vocabulary	761.4 ns	20.30 KB

License

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