Supported Languages¶

Smart Detect¶

Smart Detect is language-agnostic; it works with any language.

Transcription¶

Language support depends on which model variant you pick within each service.

Parakeet¶

The v3 model supports: Bulgarian, Croatian, Czech, Danish, Dutch, English, Estonian, Finnish, French, German, Greek, Hungarian, Italian, Latvian, Lithuanian, Maltese, Polish, Portuguese, Romanian, Russian, Slovak, Slovenian, Spanish, Swedish, Ukrainian.

Whisper¶

Whisper supports 99+ languages to varying degrees. See the full list here.

Each Whisper size (tiny / base / small / medium) also has an English-only variant (e.g. tiny.en, base.en, etc.). These are trained exclusively on English audio and are usually more accurate on English than the equivalent multilingual size. The large and turbo variants are multilingual only.

Picking a Whisper language¶

On the Transcribe Titles screen, Achew tries to preselect the language based on the book's language set in Audiobookshelf. If that metadata is missing or wrong, you'll want to specify manually.

Whisper's accuracy can drop significantly if you use the Auto language option; always pick explicitly when you know. For English books, prefer the English-only variant of a given model over the multilingual one.

AI Cleanup (LLMs)¶

Most LLM providers can handle several languages, with larger cloud services capable of dozens of languages. Each provider has its own strengths and weaknesses; you'll want to research which options will work best for your own library.

If cleanup results look wrong for a non-English book:

• Make sure that Custom/Additional Instructions are not accidentally coercing English output.
• Add a Custom Instruction like "Keep titles in the book's original language."
• Try a different model or provider.

Related¶

• Transcription
• AI Cleanup