Everything this app learns stays on this device. Export a backup file here and import it on your other phone or computer - imports merge, so learning from both devices adds together.
Talking Hurts is a free communication aid for anyone who temporarily can't speak - laryngitis, a strained voice, recovery from throat or thyroid surgery, mouth ulcers, radiotherapy, or just a day when every word costs too much. It learns as you use it: every sentence you speak becomes a one-tap suggestion the next time, organised by where you are. Install it to your home screen and it works entirely offline.
No. The phrase model, your history, your name and your saved places live in your browser's local storage. The only network use is downloading the app itself and, if you choose one, a neural voice. Backups are files you export and keep yourself.
Yes - add it to your home screen and everything works with no connection, including a downloaded neural voice. Prediction runs entirely on the device.
Out of the box it uses your device's built-in voice. You can optionally download a natural neural voice (Piper, about 60 MB, one-time) that speaks with proper sentence rhythm - much nicer for talking to people all day.
No. It is a simple free tool, not a regulated medical device or a substitute for professional speech therapy or AAC assessment. For long-term speech loss, please talk to a speech-language professional about full AAC solutions.
Free and open. Nothing tracked, nothing uploaded. Neural voices are Piper models: Male Northern English (OpenSLR 83, CC-BY-SA 4.0), Alan (Mycroft mimic3 apope), Jenny (Dioco), Alba (Edinburgh DataShare, CC-BY 4.0), and Amy and Ryan (US). Speech synthesis runs on your device with vits-web and ONNX Runtime (MIT). Made by Adam Woodland; more free tools at Adam's Arcade. Design, code and content developed with the assistance of AI (Anthropic Claude).