Good question! The short answer is, neither. There are no hardcoded rules, but the app also doesn't actively learn your personal speaking patterns over time.
All the context-awareness comes straight from the pre-trained Whisper model. Since it's a transformer network, it looks at the entire sentence context rather than translating word-by-word. For example, if you dictate a sentence about coding, it naturally knows to capitalize "Rust" and "Python" instead of writing about rusty metal and snakes.
I deliberately kept the model static. Trying to fine-tune it locally on the fly would mean I'd have to store your voice data (which kills the 100% privacy promise).
That being said, adding a custom dictionary feature, so you can feed it highly specific industry jargon right before you speak, is at the very top of my to-do list!
Let me know how it handles your vocabulary if you give the trial a spin.
Good question! The short answer is, neither. There are no hardcoded rules, but the app also doesn't actively learn your personal speaking patterns over time.
All the context-awareness comes straight from the pre-trained Whisper model. Since it's a transformer network, it looks at the entire sentence context rather than translating word-by-word. For example, if you dictate a sentence about coding, it naturally knows to capitalize "Rust" and "Python" instead of writing about rusty metal and snakes.
I deliberately kept the model static. Trying to fine-tune it locally on the fly would mean I'd have to store your voice data (which kills the 100% privacy promise).
That being said, adding a custom dictionary feature, so you can feed it highly specific industry jargon right before you speak, is at the very top of my to-do list!
Let me know how it handles your vocabulary if you give the trial a spin.