Apple's metadata system uses two separate language fields serving different purposes. Getting them confused is a common and avoidable error.
2.20.1 Two distinct fields (Apple)
**Metadata language** (release level) is the language the metadata text itself — title, artist name — is written in. It must match the language of the metadata, not the audio. **Track Audio Language** (track level) is the language actually spoken or sung in the recording, and must match the vocals. A track with an English title and Hindi vocals needs metadata language set to English and audio language set to Hindi — both fields are required and serve genuinely different functions.
2.20.2 Non-linguistic content
Purely instrumental tracks must have the Track Audio Language flagged with the ISO 639-2 code "zxx" — denoting no linguistic content. Leaving this field blank, or setting it to a spoken language for an instrumental track, is incorrect.
2.20.3 Localisations
Apple's guidance: localisations enhance discoverability. Provide artist and title metadata in your primary language, and accurate localisations for every additional language you have available — unless you specifically want the metadata to display in the primary language across all territories. For Indian-language releases, providing both native script and Roman transliteration significantly expands discoverability for diaspora audiences searching in English.
2.20.4 No side-by-side translations
Apple explicitly prohibits placing two languages in the same metadata field — "Khwaab (Dream)" combined into one title field is incorrect. Each language version belongs in its own dedicated localisation field, submitted separately.
2.20.5 Albums with multiple languages
When a release contains tracks in different languages, set the dominant language at the release level for the metadata language field. Individual tracks set their own Track Audio Language independently at the track level.
Source: MadDesk article 2.20 (2-20)