How to determine the encoding of text

Question

I received some text that is encoded, but I don't know what charset was used. Is there a way to determine the encoding of a text file using Python? How can I detect the encoding/codepage of a text file deals with C#.

Answer 1

Summary:

• cchardet for small inputs
• charset_normalizer for large inputs

When the input is random bytes (no text encoding is detected) then you probably want hex_escape

Compare some algorithms: guess-encoding-of-bytestring.py

#!/usr/bin/env python3

# guess text encoding of bytestring

# [cchardet]: https://github.com/PyYoshi/cChardet
# [faust-cchardet]: https://github.com/faust-streaming/cChardet
# [uchardet]: https://gitlab.freedesktop.org/uchardet/uchardet
# good for short strings
# fails on long strings
def guess_encoding_cchardet(bs: bytes):
    return cchardet.detect(bs).get("encoding")

# [charset_normalizer]: https://github.com/jawah/charset_normalizer
# [charset_normalizer#566]: https://github.com/jawah/charset_normalizer/issues/566
# good for long strings
# fails on short strings
#   https://github.com/jawah/charset_normalizer/issues/486
# 20x faster than chardet [charset_normalizer]
#   -> 200x slower than cchardet
# 5x slower than cchardet [charset_normalizer#566]
# benchmark versus chardet
#   https://github.com/jawah/charset_normalizer/raw/master/bin/performance.py
def guess_encoding_charset_normalizer(bs: bytes):
    match = charset_normalizer.from_bytes(bs).best()
    if match:
        return match.encoding
    return None

# [rs_chardet]: https://github.com/emattiza/rs_chardet
# 40x slower than cchardet [rs_chardet]
def guess_encoding_rs_chardet(bs: bytes):
    return rs_chardet.detect_rs_enc_name(bs)
    # return rs_chardet.detect_codec(bs).name

# [chardet]: https://github.com/chardet/chardet
# 4000x slower than cchardet [rs_chardet]
# 2000x slower than cchardet [cchardet]
def guess_encoding_chardet(bs: bytes):
    return chardet.detect(bs).get("encoding")

# [magic]: https://github.com/ahupp/python-magic
# fails on short strings
def guess_encoding_magic(bs: bytes):
    e = magic.detect_from_content(bs).encoding
    if e in ("binary", "unknown-8bit"):
        return None
    return e

# [icu]: https://github.com/unicode-org/icu
# fails on short strings
def guess_encoding_icu(bs: bytes):
    try:
        return icu.CharsetDetector(bs).detect().getName()
    except icu.ICUError:
        return None


if __name__ == "__main__":

    # test

    import random

    bytes_encoding_list = [
        ("ü".encode("latin1"), "latin1"),
        ("üü".encode("latin1"), "latin1"),
        ("üüü".encode("latin1"), "latin1"),
    ]

    for _ in range(10):
        bytes_encoding_list += [
            (random.randbytes(20), None),
        ]

    def test(guess_encoding):
        global bytes_encoding_list
        module_name = guess_encoding._name
        for input_bytes, expected_encoding in bytes_encoding_list:
            assert isinstance(input_bytes, bytes)
            # TODO better...
            guessed_encoding = guess_encoding(input_bytes)
            actual_string = None
            if guessed_encoding:
                try:
                    actual_string = input_bytes.decode(guessed_encoding)
                except Exception as exc:
                    if expected_encoding == None:
                        print(f"{module_name}: fail. found wrong encoding {guessed_encoding} in random bytes {input_bytes}")
                        continue
                    else:
                        print(f"{module_name}: FIXME failed to decode bytes: {exc}")
            if expected_encoding == None:
                # the guessed encoding can be anything -> dont compare encoding
                if guessed_encoding == None:
                    print(f"{module_name}: ok. found no encoding in random bytes {input_bytes}")
                else:
                    print(f"{module_name}: ok. found encoding {guessed_encoding} in random bytes {input_bytes} -> string {actual_string!r}")
            else:
                expected_string = input_bytes.decode(expected_encoding)
                if actual_string == expected_string:
                    print(f"{module_name}: ok. decoded {actual_string} from {guessed_encoding} bytes {input_bytes}")
                else:
                    #print(f"{module_name}: fail. actual {actual_string!r} from {guessed_encoding}. expected {expected_string!r} from {expected_encoding} bytes {input_bytes}")
                    print(f"{module_name}: fail. string: {actual_string!r} != {expected_string!r}. encoding: {guessed_encoding} != {expected_encoding}. bytes: {input_bytes}")

    for k in list(globals().keys()):
        if not k.startswith("guess_encoding_"):
            continue
        module_name = k[15:]
        module_found = False
        try:
            module = __import__(module_name)
            globals()[module_name] = module
            module_found = True
        except ModuleNotFoundError as exc:
            print(f"{module_name}: module not found. hint: pip install {module_name}")
            pass
        if module_found:
            guess_encoding = locals()[k]
            guess_encoding._name = module_name
            test(guess_encoding)

Answer 2

Some text files are aware of their encoding, but most are not. Aware:

• a text file having a BOM
• an XML file is encoded in UTF-8 or its encoding is given in the preamble
• a JSON file is always encoded in UTF-8

Not aware:

• a CSV file
• any random text file

Some encodings are versatile, i.e., they can decode any sequence of bytes, and some are not. US-ASCII is not versatile, since any byte greater than 127 is not mapped to any character. UTF-8 is not versatile since any sequence of bytes is not valid.

On the contrary, Latin-1, Windows-1252, etc. are versatile (even if some bytes are not officially mapped to a character):

>>> [b.to_bytes(1, 'big').decode("latin-1") for b in range(256)]
['\x00', ..., 'ÿ']

Given a random text file encoded in a sequence of bytes, you can't determine the encoding unless the file is aware of its encoding, because some encodings are versatile. But you can sometimes exclude non versatile encodings. All versatile encodings are still possible. The chardet modules uses the frequency of bytes to guess which encoding fits the best to the encoded text.

If you don't want to use this module or a similar one, here's a simple method:

• check if the file is aware of its encoding (BOM)
• check non versatile encodings and accept the first that can decode the bytes (ASCII before UTF-8, because it is stricter)
• choose a fallback encoding.

The second step is a bit risky if you check only a sample, because some bytes in the rest of the file may be invalid.

The code:

def guess_encoding(data: bytes, fallback: str = "iso8859_15") -> str:
    """
    A basic encoding detector.
    """
    for bom, encoding in [
        (codecs.BOM_UTF32_BE, "utf_32_be"),
        (codecs.BOM_UTF32_LE, "utf_32_le"),
        (codecs.BOM_UTF16_BE, "utf_16_be"),
        (codecs.BOM_UTF16_LE, "utf_16_le"),
        (codecs.BOM_UTF8, "utf_8_sig"),
    ]:
        if data.startswith(bom):
            return encoding

    if all(b < 128 for b in data):
        return "ascii"  # You may want to use the fallback here if data is only a sample.

    decoder = codecs.getincrementaldecoder("utf_8")()
    try:
        decoder.decode(data, final=False)
    except UnicodeDecodeError:
        return fallback
    else:
        return "utf_8"  # Not certain if data is only a sample

Remember that unversatile encodings may fail. The errors parameter of the decode method can be set to 'ignore' , 'replace' or 'backslashreplace' to avoid exceptions.

Answer 3

You can use the chardet module:

import chardet

with open (filepath , "rb") as f:
    data= f.read()
    encode=chardet.UniversalDetector()
    encode.close()
    print(encode.result)

Or you can use the chardet3 command in Linux, but it takes some time:

chardet3 fileName

Example:

chardet3 donnee/dir/donnee.csv
donnee/dir/donnee.csv: ISO-8859-1 with confidence 0.73

Answer 4

This site has Python code for recognizing ASCII, encoding with BOMs, and UTF-8 without a BOM: 8. How to guess the encoding of a document.

Here's an example. I'm on OS X.

#!/usr/bin/python

import sys

def isUTF8(data):
    try:
        decoded = data.decode('UTF-8')
    except UnicodeDecodeError:
        return False
    else:
        for ch in decoded:
            if 0xD800 <= ord(ch) <= 0xDFFF:
                return False
        return True

def get_bytes_from_file(filename):
    return open(filename, "rb").read()

filename = sys.argv[1]
data = get_bytes_from_file(filename)
result = isUTF8(data)
print(result)


PS /Users/js> ./isutf8.py hi.txt
True

Answer 5

You can use the python-magic package which does not load the whole file into memory:

import magic


def detect(
    file_path,
):
    return magic.Magic(
        mime_encoding=True,
    ).from_file(file_path)

The output is the encoding name, for example:

• iso-8859-1
• us-ascii
• utf-8

Answer 6

Using the Linux file -i command:

import subprocess

file = "path/to/file/file.txt"

encoding =  subprocess.Popen("file -bi "+file, shell=True, stdout=subprocess.PIPE).stdout

encoding = re.sub(r"(\\n)[^a-z0-9\-]", "", str(encoding.read()).split("=")[1], flags=re.IGNORECASE)

print(encoding)

Answer 7

Here is an example of reading and taking at face value a chardet encoding prediction, reading n_lines from the file in the event it is large.

chardet also gives you a probability (i.e., confidence) of its encoding prediction (I haven't looked how they come up with that), which is returned with its prediction from chardet.predict(), so you could work that in somehow if you like.

import chardet
from pathlib import Path

def predict_encoding(file_path: Path, n_lines: int=20) -> str:
    '''Predict a file's encoding using chardet'''

    # Open the file as binary data
    with Path(file_path).open('rb') as f:
        # Join binary lines for specified number of lines
        rawdata = b''.join([f.readline() for _ in range(n_lines)])

    return chardet.detect(rawdata)['encoding']

Answer 8

Depending on your platform, I just opt to use the Linux shell file command. This works for me since I am using it in a script that exclusively runs on one of our Linux machines.

Obviously, this isn't an ideal solution or answer, but it could be modified to fit your needs. In my case I just need to determine whether a file is UTF-8 or not.

import subprocess

file_cmd = ['file', 'test.txt']
p = subprocess.Popen(file_cmd, stdout=subprocess.PIPE)
cmd_output = p.stdout.readlines()

# x will begin with the file type output as is observed using 'file' command
x = cmd_output[0].split(": ")[1]
return x.startswith('UTF-8')

Answer 9

Another option for working out the encoding is to use libmagic (which is the code behind the file command). There are a profusion of Python bindings available.

The Python bindings that live in the file source tree are available as the python-magic (or python3-magic) debian package. It can determine the encoding of a file by doing:

import magic

blob = open('unknown-file', 'rb').read()
m = magic.open(magic.MAGIC_MIME_ENCODING)
m.load()
encoding = m.buffer(blob)  # "utf-8", "us-ascii", etc.

There is an identically named, but incompatible, python-magic pip package on PyPI that also uses libmagic. It can also get the encoding, by doing:

import magic

blob = open('unknown-file', 'rb').read()
m = magic.Magic(mime_encoding=True)
encoding = m.from_buffer(blob)

Answer 10

Some encoding strategies (please uncomment to taste):

#!/bin/bash
#
tmpfile=$1
echo '-- info about file file ........'
file -i $tmpfile
enca -g $tmpfile
echo 'recoding ........'
#iconv -f iso-8859-2 -t utf-8 back_test.xml > $tmpfile
#enca -x utf-8 $tmpfile
#enca -g $tmpfile
recode CP1250..UTF-8 $tmpfile

You might like to check the encoding by opening and reading the file in a form of a loop... but you might need to check the file size first:

# PYTHON
encodings = ['utf-8', 'windows-1250', 'windows-1252'] # Add more
for e in encodings:
    try:
        fh = codecs.open('file.txt', 'r', encoding=e)
        fh.readlines()
        fh.seek(0)
    except UnicodeDecodeError:
        print('got Unicode error with %s, trying different encoding' % e)
    else:
        print('opening the file with encoding:  %s ' % e)
        break

Answer 11

cchardet faster alternative to chardet

Install: pip install cchardet

Use:

import cchardet as chardet

filepath = Path(filename)

blob = filepath.read_bytes()
detection = chardet.detect(blob)

encoding = detection["encoding"]
confidence = detection["confidence"]

Answer 12

With this answer

I just want to add, for everyone's information, to install magic from the Python 3 pip:

pip install python-magic

Reference 1 Reference 2

Answer 13

A long time ago, I had this need.

Reading old code of mine, I found this:

    import urllib.request
    import chardet
    import os
    import settings

    [...]
    file = 'sources/dl/file.csv'
    media_folder = settings.MEDIA_ROOT
    file = os.path.join(media_folder, str(file))
    if os.path.isfile(file):
        file_2_test = urllib.request.urlopen('file://' + file).read()
        encoding = (chardet.detect(file_2_test))['encoding']
        return encoding

This worked for me and returned ascii

Answer 14

If you are not satisfied with the automatic tools you can try all codecs and see which codec is right manually.

all_codecs = ['ascii', 'big5', 'big5hkscs', 'cp037', 'cp273', 'cp424', 'cp437', 
'cp500', 'cp720', 'cp737', 'cp775', 'cp850', 'cp852', 'cp855', 'cp856', 'cp857', 
'cp858', 'cp860', 'cp861', 'cp862', 'cp863', 'cp864', 'cp865', 'cp866', 'cp869', 
'cp874', 'cp875', 'cp932', 'cp949', 'cp950', 'cp1006', 'cp1026', 'cp1125', 
'cp1140', 'cp1250', 'cp1251', 'cp1252', 'cp1253', 'cp1254', 'cp1255', 'cp1256', 
'cp1257', 'cp1258', 'euc_jp', 'euc_jis_2004', 'euc_jisx0213', 'euc_kr', 
'gb2312', 'gbk', 'gb18030', 'hz', 'iso2022_jp', 'iso2022_jp_1', 'iso2022_jp_2', 
'iso2022_jp_2004', 'iso2022_jp_3', 'iso2022_jp_ext', 'iso2022_kr', 'latin_1', 
'iso8859_2', 'iso8859_3', 'iso8859_4', 'iso8859_5', 'iso8859_6', 'iso8859_7', 
'iso8859_8', 'iso8859_9', 'iso8859_10', 'iso8859_11', 'iso8859_13', 
'iso8859_14', 'iso8859_15', 'iso8859_16', 'johab', 'koi8_r', 'koi8_t', 'koi8_u', 
'kz1048', 'mac_cyrillic', 'mac_greek', 'mac_iceland', 'mac_latin2', 'mac_roman', 
'mac_turkish', 'ptcp154', 'shift_jis', 'shift_jis_2004', 'shift_jisx0213', 
'utf_32', 'utf_32_be', 'utf_32_le', 'utf_16', 'utf_16_be', 'utf_16_le', 'utf_7', 
'utf_8', 'utf_8_sig']

def find_codec(text):
    for i in all_codecs:
        for j in all_codecs:
            try:
                print(i, "to", j, text.encode(i).decode(j))
            except:
                pass

find_codec("The example string which includes ö, ü, or ÄŸ, ö")

This script creates at least 9409 lines of output. So, if the output cannot fit to the terminal screen try to write the output to a text file.

Answer 15

EDIT: chardet seems to be unmantained but most of the answer applies. Check https://pypi.org/project/charset-normalizer/ for an alternative

Correctly detecting the encoding all times is impossible.

(From chardet FAQ:)

However, some encodings are optimized for specific languages, and languages are not random. Some character sequences pop up all the time, while other sequences make no sense. A person fluent in English who opens a newspaper and finds “txzqJv 2!dasd0a QqdKjvz” will instantly recognize that that isn't English (even though it is composed entirely of English letters). By studying lots of “typical” text, a computer algorithm can simulate this kind of fluency and make an educated guess about a text's language.

There is the chardet library that uses that study to try to detect encoding. chardet is a port of the auto-detection code in Mozilla.

You can also use UnicodeDammit. It will try the following methods:

• An encoding discovered in the document itself: for instance, in an XML declaration or (for HTML documents) an http-equiv META tag. If Beautiful Soup finds this kind of encoding within the document, it parses the document again from the beginning and gives the new encoding a try. The only exception is if you explicitly specified an encoding, and that encoding actually worked: then it will ignore any encoding it finds in the document.
• An encoding sniffed by looking at the first few bytes of the file. If an encoding is detected at this stage, it will be one of the UTF-* encodings, EBCDIC, or ASCII.
• An encoding sniffed by the chardet library, if you have it installed.
• UTF-8
• Windows-1252

Answer 16

This might be helpful

from bs4 import UnicodeDammit
with open('automate_data/billboard.csv', 'rb') as file:
   content = file.read()

suggestion = UnicodeDammit(content)
suggestion.original_encoding
#'iso-8859-1'

Answer 17

# Function: OpenRead(file)

# A text file can be encoded using:
#   (1) The default operating system code page, Or
#   (2) utf8 with a BOM header
#
#  If a text file is encoded with utf8, and does not have a BOM header,
#  the user can manually add a BOM header to the text file
#  using a text editor such as notepad++, and rerun the python script,
#  otherwise the file is read as a codepage file with the 
#  invalid codepage characters removed

import sys
if int(sys.version[0]) != 3:
    print('Aborted: Python 3.x required')
    sys.exit(1)

def bomType(file):
    """
    returns file encoding string for open() function

    EXAMPLE:
        bom = bomtype(file)
        open(file, encoding=bom, errors='ignore')
    """

    f = open(file, 'rb')
    b = f.read(4)
    f.close()

    if (b[0:3] == b'\xef\xbb\xbf'):
        return "utf8"

    # Python automatically detects endianess if utf-16 bom is present
    # write endianess generally determined by endianess of CPU
    if ((b[0:2] == b'\xfe\xff') or (b[0:2] == b'\xff\xfe')):
        return "utf16"

    if ((b[0:5] == b'\xfe\xff\x00\x00') 
              or (b[0:5] == b'\x00\x00\xff\xfe')):
        return "utf32"

    # If BOM is not provided, then assume its the codepage
    #     used by your operating system
    return "cp1252"
    # For the United States its: cp1252


def OpenRead(file):
    bom = bomType(file)
    return open(file, 'r', encoding=bom, errors='ignore')


#######################
# Testing it
#######################
fout = open("myfile1.txt", "w", encoding="cp1252")
fout.write("* hi there (cp1252)")
fout.close()

fout = open("myfile2.txt", "w", encoding="utf8")
fout.write("\u2022 hi there (utf8)")
fout.close()

# this case is still treated like codepage cp1252
#   (User responsible for making sure that all utf8 files
#   have a BOM header)
fout = open("badboy.txt", "wb")
fout.write(b"hi there.  barf(\x81\x8D\x90\x9D)")
fout.close()

# Read Example file with Bom Detection
fin = OpenRead("myfile1.txt")
L = fin.readline()
print(L)
fin.close()

# Read Example file with Bom Detection
fin = OpenRead("myfile2.txt")
L =fin.readline() 
print(L) #requires QtConsole to view, Cmd.exe is cp1252
fin.close()

# Read CP1252 with a few undefined chars without barfing
fin = OpenRead("badboy.txt")
L =fin.readline() 
print(L)
fin.close()

# Check that bad characters are still in badboy codepage file
fin = open("badboy.txt", "rb")
fin.read(20)
fin.close()

Answer 18

If you know the some content of the file you can try to decode it with several encoding and see which is missing. In general there is no way since a text file is a text file and those are stupid ;)

Answer 19

It is, in principle, impossible to determine the encoding of a text file, in the general case. So no, there is no standard Python library to do that for you.

If you have more specific knowledge about the text file (e.g. that it is XML), there might be library functions.