Say What

Writing by voice works well only when software understands the voice in front of it. Say What is a local Greek dictation tool being developed for people with speech impediments, breathing disorders and other disabilities that make keyboards or general-purpose dictation systems difficult. It is intended for practical and long-form work: composing letters, leaving messages and narrating books one phrase at a time.

The prototype listens for the speaker’s natural pauses instead of treating them as mistakes. Silero voice-activity detection separates speech into breath groups; faster-whisper transcribes each group in Greek; the preceding passage returns as context so new phrases can be joined into editable prose. Pause length and model size remain adjustable, existing recordings can be transcribed, and the text autosaves locally. Audio stays on the computer.

We built and tested the pipeline in Python with Tkinter, sounddevice, CTranslate2 and Whisper. The current result is a research prototype: useful enough to test with real speech, but not yet a finished assistive product. These images are concept studies for how its interface, and a possible dedicated device, could develop.