Voice Input for Expenses
Letting users speak transactions — and cleaning speech-to-text for financial accuracy.
People abandon expense apps because typing NPR 80 for tea feels ridiculous. Voice is the obvious fix — until speech-to-text mangles numbers and merchant names.
BolKharcha treats voice as an input pipeline, not a magic mic button. Capture audio, transcribe, normalize, interpret, confirm. Each stage can fail gracefully.
Number normalization is a project by itself. Spoken “forty five hundred” and typed “4500” must land in the same decimal world. Currency words need local awareness.
Privacy expectations matter. Users should know when recording starts and ends. Clear affordances beat hidden always-listening fantasies.
Offline and noisy environments are real. Provide keyboard fallback immediately. Voice is acceleration, not a gate.
The best voice UX ends with an editable summary. Let users fix one field without re-recording the whole sentence.
When voice works, daily logging finally matches daily life: quick, spoken, done.