Hold a key, speak, let go. Finished text lands where your cursor is, in the language you were speaking.
Mac only*Live words appear while you talk. When you let go, a larger model replaces them with a better result in about a second. It is tuned for accented English, Indian English in particular, and for the names and figures other dictation tools guess at.
Pick from the menu bar and the text comes out in that script. Hinglish is its own option, written in English letters: "Kal ka plan kya hai? Mujhe document bhej do." Switch mid-day between 16 languages without touching a setting.
Record your microphone and whatever the Mac is playing. Name a voice once and it is recognised in every later call. When it is not sure, it says so instead of pretending.
Audio goes to the transcription model only while you hold the key, and is discarded once the text is typed unless you choose to keep it for retries. Voices are stored as numbers, never audio. Nothing you say is used for training.