what is this?
create dadaist poetry by transcribing a stream of mangled speech with a
tiny speech-to-text model running in your web browser. as you draw on the canvas, you will hear the mangled speech going into the model, and the model's attempts to transcribe it will appear. you can assemble these surreal phrases into poetry by dragging them into place.
controls
- left click/one touch and drag: mangle audio and generate text.
- right click/two touch and drag: move text. press space or tap with a third finger while dragging to delete.
- q/four touch: open menu. you can record audio and adjust settings.
privacy
the model is running in your web browser, so nothing leaves your computer. all of the inferences are happening here; all of the audio you record stays here.
safety
speech-to-text models are designed to transcribe any and all words that people say––good or bad. people sometimes say things we don't agree with, or things that are harmful or offensive. since this project feeds the model weird audio, you should expect weird output. in rare cases it may be offensive. i assume no responsibility for the words that appear while you use this project.