The microphone button on the home screen records an expense from speech. Say the amount and the category: "two hundred pesos groceries", "fifteen hundred fuel".
Recognition works in one language at a time — browsers cannot detect the language on the fly. The dictation language is chosen in the recording sheet and remembered. If you spoke another language and it did not understand, it offers to try again in that one.
The recognition is done by the browser. The app stores only the parsed result — the amount, the category, and your wording if you agreed to remember it. No audio is sent anywhere or kept.