93

There are many plain text files which were encoded in variant charsets.

I want to convert them all to UTF-8, but before running iconv, I need to know its original encoding. Most browsers have an Auto Detect option in encodings, however, I can't check those text files one by one because there are too many.

Only having known the original encoding, I then can convert the texts by iconv -f DETECTED_CHARSET -t utf-8.

Is there any utility to detect the encoding of plain text files? It DOES NOT have to be 100% perfect, I don't mind if there're 100 files misconverted in 1,000,000 files.

Improve this question

11 Answers 11

Reset to default
80

Try the chardet Python module, which is available on PyPI:

pip install chardet

Then run chardetect myfile.txt.

Chardet is based on the detection code used by Mozilla, so it should give reasonable results, provided that the input text is long enough for statistical analysis. Do read the project documentation.

As mentioned in comments it is quite slow, but some distributions also ship the original C++ version as @Xavier has found in https://superuser.com/a/609056. There is also a Java version somewhere.

Improve this answer
6
  • 5
    Yes, and it's already packaged as python-chardet in Ubuntu universe repo.
    – Lenik
    Jun 25, 2011 at 6:21
  • If it wasn't a perfect guess, chardet will still give the most correctly guess, like ./a.txt: GB2312 (confidence: 0.99). Compared to Enca which just failed and report 'Unrecognized encoding'. However, sadly enough, chardet runs very slow.
    – Lenik
    Jun 25, 2011 at 6:48
  • 2
    @谢继雷: Have it run overnight or something like that. Charset detection is a complicated process. You could also try the Java-based jChardet or ... the original chardet is part of Mozilla, but only C++ source is available, no command-line tool.
    – u1686_grawity
    Jun 25, 2011 at 12:13
  • 2
    Regarding speed: running chardet <(head -c4000 filename.txt) was much faster and equally successful for my use-case. (in case it's not clear this bash syntax will send only the first 4000 bytes to chardet)
    – ndemou
    Dec 26, 2015 at 19:32
  • 2
    @ndemou I have chardet==3.0.4, and the command line tool's actual executable name is chardetect not chardet.
    – Devy
    Mar 26, 2018 at 14:26
42

I would use this simple command:

encoding=$(file -bi myfile.txt)

Or if you want just the actual character set (like utf-8):

encoding=$(file -b --mime-encoding myfile.txt)
Improve this answer
5
  • 8
    Unfortunately, file only detects encodings with specific properties, such as UTF-8 or UTF-16. The rest -- oldish ISO8859 or their MS-DOS and Windows correspondents -- are listed as "unknown-8bit" or something similar, even for files which chardet detects with 99% confidence.
    – u1686_grawity
    Oct 28, 2011 at 19:09
  • 9
    file showed me iso-8859-1
    – cweiske
    Mar 30, 2012 at 7:22
  • What if the extension is lying?
    – james.garriss
    Oct 3, 2014 at 13:24
  • 3
    @james.garriss: file extension has nothing to do with its (text) content encoding.
    – MestreLion
    Nov 28, 2014 at 12:18
  • In case some noob like me gets confused as to how to retrieve the value that we're storing in "encoding", simply type: "echo $encoding". Alternatively, you can use the part inside the parentheses by itself, without storing any variables: "file -bi myfile.txt".
    – gignu
    Mar 14, 2021 at 22:30
34

On Debian-based Linux, the uchardet package (Debian / Ubuntu) provides a command line tool. See below the package description:

 universal charset detection library - cli utility
 .
 uchardet is a C language binding of the original C++ implementation
 of the universal charset detection library by Mozilla.
 .
 uchardet is a encoding detector library, which takes a sequence of
 bytes in an unknown character encoding without any additional
 information, and attempts to determine the encoding of the text.
 .
 The original code of universalchardet is available at
 http://lxr.mozilla.org/seamonkey/source/extensions/universalchardet
 .
 Techniques used by universalchardet are described at
 http://www.mozilla.org/projects/intl/UniversalCharsetDetection.html
Improve this answer
3
  • 3
    Thanks! From the project's homepage it wasn't obvious to me that there was a CLI included. It's also available on OS X when installing uchardet via Homebrew.
    – Stefan Schmidt
    Jul 6, 2013 at 14:47
  • 1
    I was a little confused at first because a ISO 8859-1 document was falsely identified as Windows-1252 but in the printable range Windows-1252 is a superset of ISO 8859-1 so conversion with iconv works fine.
    – Stefan Schmidt
    Jul 6, 2013 at 14:56
  • Hi @StefanSchmidt, charset detection is based on corpus statistics. In countries like Brazil the old official standard encode (before UTF8) was ISO-8859-1... But it is a "Microsoft country", so the most popular in any statistics is the Microsoft's distortion of ISO (against competition), that is Windows-1252. It is a side effect of the technological colonialism in a country: weak sovereignty, weak official standards and strong Microsoft lobbying. Even now (2021 is 1 decade after universal adoption!) it is dominant in TXT and CSV files.
    – Peter Krauss
    Apr 9, 2021 at 10:42
17

For Linux, there is enca and for Solaris you can use auto_ef.

Improve this answer
3
  • Enca seems too strict for me: enca -d -L zh ./a.txt failed with message ./a.txt: Unrecognized encoding Failure reason: No clear winner. As @grawity mentioned, chardet is more lax, however it's yet too slow.
    – Lenik
    Jun 25, 2011 at 7:06
  • 13
    Enca completely fails the "actually does something" test.
    – Michael Wolf
    Mar 1, 2012 at 18:59
  • 2
    uchardet failed (detected CP1252 instead of the actual CP1250), but enca worked fine. (single example, hard to generalize...)
    – Palo
    Nov 16, 2015 at 20:52
3

For those regularly using Emacs, they might find the following useful (allows to inspect and validate manually the transfomation).

Moreover I often find that the Emacs char-set auto-detection is much more efficient than the other char-set auto-detection tools (such as chardet).

(setq paths (mapcar 'file-truename '(
 "path/to/file1"
 "path/to/file2"
 "path/to/file3"
)))

(dolist (path paths)
  (find-file path)
  (set-buffer-file-coding-system 'utf-8-unix)
  )

Then, a simple call to Emacs with this script as argument (see the "-l" option) does the job.

Improve this answer
2

Mozilla has a nice codebase for auto-detection in web pages:
http://lxr.mozilla.org/seamonkey/source/extensions/universalchardet/src/

Detailed description of the algorithm:
http://www-archive.mozilla.org/projects/intl/UniversalCharsetDetection.html

Improve this answer
1

UTFCast is worth a try. Didn't work for me (maybe because my files are terrible) but it looks good.

http://www.addictivetips.com/windows-tips/how-to-batch-convert-text-files-to-utf-8-encoding/

Improve this answer
1

Getting back to chardet (python 2.?) this call might be enough:

python -c 'import chardet,sys; print chardet.detect(sys.stdin.read())' < file
{'confidence': 0.98999999999999999, 'encoding': 'utf-8'}

Though it's far from perfect....

echo "öasd" | iconv -t ISO-8859-1 | python -c 'import chardet,sys; print chardet.detect(sys.stdin.read())'
{'confidence': 0.5, 'encoding': 'windows-1252'}
Improve this answer
1

isutf8 (from the moreutils package) did the job

Improve this answer
2
  • 2
    How? This answer isn't really helpful.
    – user201262
    Oct 28, 2015 at 19:02
  • 2
    It's not exactly was asked, but is a useful tool. If the file is valid UTF-8, the exit status is zero. If the file is not valid UTF-8, or there is some error, the exit status is non-zero.
    – ton
    Feb 16, 2016 at 17:34
1

If you're unhappy with chardet like I am because it doesn't properly recognize some encodings, try out detect-file-encoding-and-language. I've found it to be a lot more reliable than chardet.

1. Make sure you have Node.js and NPM installed. You can install them like this:

$ sudo apt install nodejs npm

2. Install detect-file-encoding-and-language:

$ npm install -g detect-file-encoding-and-language

3. Now you can use it to detect the encoding:

$ dfeal "/home/user name/Documents/subtitle file.srt"

It'll return an object with the detected encoding, language, and a confidence score.

Improve this answer
0

Also in case you file -i gives you unknown

You can use this php command that can guess charset like below :

In php you can check like below :

Specifying encoding list explicitly : 

php -r "echo 'probably : ' . mb_detect_encoding(file_get_contents('myfile.txt'), 'UTF-8, ASCII, JIS, EUC-JP, SJIS, iso-8859-1') . PHP_EOL;"

More accurate "mb_list_encodings": 

php -r "echo 'probably : ' . mb_detect_encoding(file_get_contents('myfile.txt'), mb_list_encodings()) . PHP_EOL;"

Here in first example, you can see that i put a list of encodings (detect list order) that might be matching. To have more accurate result you can use all possible encodings via : mb_list_encodings()

Note mb_* functions require php-mbstring

apt-get install php-mbstring

See answer : https://stackoverflow.com/a/57010566/3382822

Improve this answer

You must log in to answer this question.

Not the answer you're looking for? Browse other questions tagged .