https://whisperstream.io

Hi HN, I built Whisperstream, a free and open-source dictation app for Windows.

You've seen a dozen of these, so what makes it different from all the other dictation apps?

- Local cleanup: We're the only free and open source solution on Windows with a finetuned cleanup model. Stream-1 Miniis a Qwen 0.6B finetune for removing filler and applying spoken corrections. Superwhisper is the only Windows app that offers this, is paid, and in our benchmark we outperform their S1 mini with 36% less failures. Benchmark: https://whisperstream.io/stream-1-mini-vs-s1-mini - Agent Speak: Hear or read response summaries from Claude Code or Codex, and speak back directly. Expand to see the full response without switching contexts. The TTS uses Kokoro for highly realistic voices. - Smart Dictionary: Most local dictation apps don't offer dictionary support when using their fine-tuned cleanup model. We trained it into our finetune and use phonetic matching so we only include similar words for boosting.

It's GPL-3.0, completely free and now has a Mac version too.

Demo: https://whisperstream.io/press/demos/launch-2.0-light.gif Download: https://whisperstream.io/ Source: https://github.com/lanreal-io/whisperstream Cleanup model: https://huggingface.co/whisperstream/stream-1-mini