Whisperstream
On-device dictation for Windows and Mac with broad app compatibility.
Description
On-device dictation for Windows and Mac with broad app compatibility. Local speech recognition, free and open source, with no subscription and no cloud transcription.
About Whisperstream
What is Whisperstream?
Whisperstream is an open-source, on-device dictation application for Windows and Mac that runs local speech recognition without cloud transcription.
What Whisperstream does
Whisperstream transcribes speech locally across 39 languages using models including NVIDIA Parakeet and Qwen3 ASR. It delivers text to focused fields in applications such as Outlook, Word, Slack, VS Code, browsers, and terminals via clipboard paste or simulated keystrokes. The app offers push-to-talk hotkeys, system-volume ducking, local encrypted history, and optional local AI cleanup with Stream-1 Mini.
Who Whisperstream is for
Users on Windows 10, Windows 11, or Apple silicon Mac systems who want local voice dictation and speech-to-text without sending audio to cloud servers.
How Whisperstream works
Users hold a configurable push-to-talk hotkey (Right Shift by default) to dictate speech, which on-device models transcribe locally. Optional Stream-1 Mini cleans up filler words and spoken self-corrections before delivering text to the active application.
Key features
- On-device transcription across 39 languages using NVIDIA Parakeet, Qwen3 ASR, or Cohere Transcribe with no audio sent to cloud servers.
- Stream-1 Mini local cleanup model removing filler words and applying spoken self-corrections entirely on compatible hardware.
- Configurable push-to-talk or toggle dictation shortcut defaulting to Right Shift with no administrator elevation needed.
- Cross-application text delivery into Outlook, Word, Slack, VS Code, terminals, and web browsers via clipboard or simulated typing.
- Encrypted local transcript history allowing users to search past entries, replay saved audio, and copy text.
- System volume ducking that lowers or mutes system volume while you dictate.
- Agent Speak integration linking voice dictation to local coding agents like Claude Code and Codex.
Facts from whisperstream.io, checked 13 November 2026
- Free and open-source dictation toolkit for Windows and Apple Silicon Mac.
- Uses on-device speech models including NVIDIA Parakeet and Qwen3 ASR across 39 languages.
- Optional Stream-1 Mini cleanup model downloads at about 460 MB for local text cleanup.
- Connects to coding agents such as Claude Code and Codex via Agent Speak.
Written from Whisperstream's own website (Homepage, Features, Docs). The maker can claim this listing to edit it.
Categories
Frequently asked questions
Does Whisperstream send audio to cloud servers?
No. Transcription runs entirely on your Windows PC or Apple Silicon Mac using on-device models, with no audio uploaded, streamed, or stored on remote servers.
Which platforms and operating systems are supported?
Whisperstream supports Windows 10 and Windows 11 on x64 computers, as well as Mac computers with Apple silicon running macOS 14 Sonoma or later.
How many languages can Whisperstream transcribe?
Whisperstream transcribes 39 languages, utilizing NVIDIA Parakeet for 25 languages and Qwen3 ASR for 14 additional languages.
What does Stream-1 Mini do?
Stream-1 Mini is an optional local cleanup model that removes filler words, fixes punctuation, and applies spoken corrections without sending text off your computer.
Listed from the product's public website. Own this product? Claim this listingRequest removal