WhisprStream
A local dictation app for Apple silicon Macs that types speech into other apps, handles mixed-language sentences, and works offline after model downloads.
WhisprStream turns speech into text at your cursor in other Mac apps. It's for people who dictate notes, email, or coding prompts and want speech processing to stay on their own computer.
- Mixed-language dictation: Switch languages within a sentence, including speech that mixes more than two languages.
- Live transcription: A floating display shows text as you speak, with live previews even when the speech model doesn't support native streaming.
- Custom vocabulary: Supply names and technical terms to help the app recognize words specific to your work.
- Voice shortcuts: Expand a spoken phrase into saved text, such as a signature, address, URL, or reusable reply.
The app runs Qwen3-ASR locally through MLX and offers 0.6B and 1.7B models. The speech engine and model require an initial download; afterward, dictation works offline without an account, server, or audio upload. You can customize the keyboard shortcut used to control recording.
WhisprStream is open source under the MIT license. It requires macOS 14 or later and an Apple silicon Mac, with approximately 4 GB of free storage for the engine and recommended 0.6B model. The 1.7B model is best suited to Macs with at least 16 GB of unified memory. Public downloads are self-signed and aren't notarized by Apple, so macOS may require approval before the first launch.
Logotyper
