How many languages does AI transcription support?
Short answer
The best models (Whisper, AssemblyAI, DeepScript) cover 99 languages — but quality ranges from excellent for the top 10 to barely usable for rare languages.
The answer depends on what "support" means. Most modern models claim 99+ languages, but quality is anything but uniform.
Tier 1 — excellent (WER 3-7%) English, Spanish, French, German, Italian, Portuguese, Dutch, Japanese, Mandarin Chinese, Russian. Each has tens of thousands of hours of training data. Modern AI here is near-human on clean audio.
Tier 2 — very good (WER 7-12%) Polish, Korean, Turkish, Standard Arabic, Swedish, Danish, Finnish, Norwegian, Czech, Greek, Hebrew, Ukrainian. Solid for most use cases, light post-editing helps.
Tier 3 — usable (WER 12-20%) Many Indian languages (Hindi, Tamil, Telugu, Bengali), Vietnamese, Thai, Indonesian, Southeast European languages (Croatian, Serbian, Slovak, Slovenian, Bulgarian, Romanian, Hungarian), Baltic languages (Lithuanian, Latvian, Estonian).
Tier 4 — limited (WER > 20%) Rare languages with little training material — many African languages, broad Swahili dialects beyond standard, Amharic, small indigenous languages. Heavy regional accents land here too (Swiss German, broad Bavarian, broad Glaswegian).
Language detection (auto-detect) Most models detect the language from the first 30 seconds of audio. Works reliably for the top 30, falls back to wrong language on rare languages or short clips. If you know the language, set it explicitly — beats auto-detect every time.
Multilingual recordings For recordings with language switches (e.g. a German interview with an English quote inserted) there are two modes: - Preserve languages: model transcribes each segment in its original language. Tier-1 models can do this, many others can't. - Translate: model translates everything into a target language. Whisper offers `task=translate` to push everything to English.
DeepScript languages We support 99 languages. The UI shows a dropdown with all of them; on the API it's the `language` parameter as ISO 639-1 (`de`, `en`, `fr`, `it`, `es`, `pl`, `nl`, …). For DACH markets, Premium with DACH tuning is especially relevant — see the Swiss German question.
Practical note "99 languages" is a marketing number — it tells you nothing about how well any single language works. If you need a rarely supported language, ask the provider for a concrete sample and verify the output yourself. Don't trust a checkbox table.
Related questions
Still have a question?
Three transcriptions free to try. Or drop us a line — we answer within 24 hours, compliance questions included.