0.16.1 · September 2026
- Model downloads that don't give up. If the speech model can't be downloaded, AlphaDictate now tries again on its own and falls back to a third server, and it trusts the same secure connections your browser does. A model you download from the Model page starts working right away, instead of the old error staying on screen.
0.16.0 · September 2026
- Lighter when idle. Between dictations, AlphaDictate now frees the up to 2 GB the speech model takes and loads it again while you speak, and it no longer uses any CPU while it waits. Prefer it always ready? Turn on Model → Keep the model loaded between dictations.
- Several triggers per action. Give start and stop, or any other action, a keyboard shortcut, a mouse button, or both.
- No more slow first dictation. After an install or a graphics driver update, the first dictation could take 15–20 seconds while the graphics card prepared itself. That now happens while the model loads.
- Fairer microphone advice. Each microphone is judged on its own recordings, and 16 kHz USB microphones are no longer mistaken for Bluetooth headsets.
- Vocabulary hints that stick. The first word of your vocabulary hint no longer comes out garbled.
- Mouse buttons in your language. Mouse button names now follow the app's language.
- Download problems explained in your language. If a speech model can't be downloaded, the reason is now shown in the app's language, not only in English.
- Tidier History. A recording you are playing stops when you switch to another page.
- Steadier settings. Rows no longer shift when you flip a switch or change the pill's size.