AudioWhisper
A macOS dictation app that transcribes speech locally with WhisperKit or Parakeet-MLX, or sends audio to OpenAI Whisper or Google Gemini.
AudioWhisper is an open source macOS menu bar app for people who want to dictate into other apps or turn recorded audio into text. It combines offline transcription with optional cloud services, so you can choose where your audio gets processed. The project uses the MIT license.
WhisperKit runs on-device through CoreML, with Whisper models including Tiny, Base, Small and Large Turbo. Parakeet-MLX provides another local option with English and multilingual models. Both work offline once their models are downloaded and don't require API keys. Local Whisper also runs on Intel Macs, though it's slower; Parakeet and local MLX text correction require Apple Silicon. The app requires macOS Sonoma or later.
Global shortcuts, push-to-talk and a windowless recording mode make it suited to dictation while you're working in another app. It copies transcripts to the clipboard and can paste them automatically, then return focus to your previous app. It also transcribes existing audio files. Optional text correction fixes typos and punctuation and removes filler words, with editable categories for contexts such as coding and email.
Correction can run locally through MLX or use a cloud provider. OpenAI Whisper and Google Gemini transcription require internet access and API keys, and send audio to those services. Local transcription keeps audio on your Mac. The app collects no analytics, stores API keys in macOS Keychain, and can keep a searchable local transcript history with retention controls. Its usage dashboard tracks word counts, dictation speed and estimated typing time saved.
