Finnish Subtitle Generator

Finnish is not related to the languages around it, and models trained mostly on Indo-European speech treat it badly. Fifteen grammatical cases, expressed as suffixes on the noun, mean that one word has dozens of forms. Vowel harmony decides which vowels those suffixes can contain. Consonant gradation changes the stem itself — kukka becomes kukan. A transcript has to know all of this to write the ending correctly, because the audio for two endings is often almost identical.

Length matters as much as letters. Finnish distinguishes short and long vowels and consonants, and writing one letter where there should be two produces a different word. Tuli is fire, tuuli is wind, tulli is customs. The transcript writes the length that was spoken, which is a detail most general tools blur.

Finally there is the gap between spoken and written Finnish, which is wider than in most languages. Puhekieli — everyday spoken Finnish — shortens pronouns, drops endings and uses forms that never appear in print. The transcript follows the speaker by default, and if you want subtitles in formal written Finnish the editor lets you standardise before you export.