"MARC (ISO 2709)" records could be encoded in two different character coding schemes: MARC-8 or UCS/Unicode.
Use yaz-marcdump to convert the encoding of MARC records. Specify the encoding with options -f and -t. With option -l you can set the character coding scheme in the MARC leader position 09.
$ yaz-marcdump -f MARC-8 -t UTF-8 -o marc -l 9=97 marc21.raw > marc21.utf8.raw
A conversion from UTF-8 to MARC-8 is not recommended, because it could be lossy.
Unicode provides single code points for many characters that could be viewed as combinations of two or more characters, e.g. German umlauts:
| Composed/NFC | Decomposed/NFD |
|---|---|
| ä (Latin Small Letter A with Diaeresis U+00E4) | a (Latin Small Letter A U+0061) + ◌̈ (Combining Diaeresis U+0308) |
With the command-line utility uconv you can transliterate data between different Unicode normalization forms:
$ uconv -x NFC marc21.nfd.xml > marc21.nfc.xml
$ uconv -x NFD marc21.nfc.xml > marc21.nfd.xml
You should only normalize "MARC XML" data, as the normalization of "MARC (ISO 2709)" would result in corrupted records, due to changed field lengths.