What’s New · September 2026

Translate locally. Dictate in more languages.

Diktator 2.1 expands private speech-to-text on Mac with a selectable translation target, two local translation paths, and stronger multilingual processing.

The essentials in Diktator 2.1

  • Choose a translation target and compare translation with the local speech model or Apple Translation.
  • Arabic and Turkish now have prepared interface localization and dedicated post-processing profiles.
  • Chinese and Japanese receive stronger language-specific cleanup and mixed-script protection.
  • The dictation picker now offers the complete model language catalog, beyond the 13 prepared interface languages.
  • Speech recognition and local-model translation remain on your Mac.

Detailed changes

A closer look at the user-facing changes in this release.

Choose the translation language

Translate dictated speech to a selected target from the full language catalog. Local-model translation is experimental: results depend on the source language, target language, and model.

Two translation paths

Use the local speech model, or choose Apple Translation. Diktator keeps local-model translation on your Mac; Apple Translation is local only when the required language resources and On-Device Mode are enabled in macOS.

More dictation languages

The dictation picker now exposes the complete model language catalog instead of limiting choices to the 13 prepared interface languages.

Arabic and Turkish

Arabic and Turkish now have prepared interface localization and dedicated post-processing profiles, bringing the prepared language set to 13.

Improved Chinese and Japanese

Language-specific plausibility checks and post-processing improve Chinese and Japanese output while protecting legitimate mixed-script text.

Mixed-language segmentation

Short segments can detect language independently, with language-specific cleanup and less context leakage from one script into the next.

Local-first privacy

Your dictation stays on your Mac.

Speech recognition and local-model translation do not send dictated audio or recognized text to a cloud transcription service. If you choose Apple Translation, macOS controls whether that translation uses downloaded on-device resources or Apple’s online service.

Diktator 2.1 · Build 61

Released September 1, 2026 for Apple silicon Macs running macOS 26 or later.

View the complete version history