Tested caption languages for Adobe Premiere Pro.
Caption Plug's bundled local Whisper model is multilingual. These are the languages we tested on real held-out speech before publishing a page—not every language the underlying model can theoretically recognize.
14
3
v1.1.3
Promising on this small test
English · English
Mean character error rate: 1.0% · Strong smoke-test result
Latin · LTR · tested 2026-08-02
Español · Spanish
Mean character error rate: 1.6% · Strong smoke-test result
Latin · LTR · tested 2026-08-02
Bahasa Indonesia · Indonesian
Mean character error rate: 8.7% · Strong smoke-test result
Latin · LTR · tested 2026-08-02
Deutsch · German
Mean character error rate: 7.4% · Strong smoke-test result
Latin · LTR · tested 2026-08-02
Türkçe · Turkish
Mean character error rate: 9.4% · Strong smoke-test result
Latin · LTR · tested 2026-08-02
Useful, with a deliberate correction pass
Português · Portuguese
Mean character error rate: 11.3% · Usable with close review
Latin · LTR · tested 2026-08-02
العربية · Arabic
Mean character error rate: 21.1% · Usable with close review
Arabic · RTL · tested 2026-08-02
Français · French
Mean character error rate: 12.4% · Usable with close review
Latin · LTR · tested 2026-08-02
日本語 · Japanese
Mean character error rate: 23.6% · Usable with close review
Japanese (Kanji, Hiragana, Katakana) · LTR · tested 2026-08-02
한국어 · Korean
Mean character error rate: 15.6% · Usable with close review
Hangul · LTR · tested 2026-08-02
Tiếng Việt · Vietnamese
Mean character error rate: 19.3% · Usable with close review
Latin · LTR · tested 2026-08-02
Polski · Polish
Mean character error rate: 10.9% · Usable with close review
Latin · LTR · tested 2026-08-02
Български · Bulgarian
Mean character error rate: 16.4% · Usable with close review
Cyrillic · LTR · tested 2026-08-02
Publish only after native review
A failed test does not become an SEO page
Hindi is offered by the underlying model, but our August 2 smoke test returned the wrong script on the tested speech and a 90.1% mean character error rate. It is not approved for a public language landing page. The failed result remains in the downloadable benchmark for accountability.
Evidence, not an accuracy guarantee
Three clips can expose obvious failures and script problems; they cannot represent every speaker, microphone, dialect, or production. Every automatic transcript still needs a human correction pass before delivery.
Read the complete method and thresholds