Live Captions (Real-time Speech to Text)

About the Live Captions (Real-time Speech to Text)

Turn speech into large, readable captions on your screen as it is spoken: a lecture, a meeting in the room, a conversation, or the sound of a video playing in another browser tab. Make the captions as big as you need, white or yellow on black, two to five lines at a time, or full screen — and add a live translation under each line. When you stop, the transcript is there with a time on every line, to keep as text or SRT subtitles.

Captions are made on your own device. Your browser’s recognition shows words as they are heard, and runs on the device where your browser can do that for your language; its online speech service is used only if you switch it on. An on-device model works in every modern browser and for a tab’s sound: it shows each phrase a second or two after it is said. MySmartCoPilot never receives your audio. Without a pass, captions run for five minutes at a time and the transcript shows its first half.

Common uses

• Following a lecture, a talk or a meeting when you are deaf or hard of hearing, or in a noisy room.
• Captions for a video or a call in another browser tab that has none.
• A rough transcript of what was said, kept as text or subtitles.
• Reading speech in another language with a live translation under it.

The two ways to caption

• Your browser’s recognition shows the words as they are heard and corrects them as the sentence goes on. Chrome and Edge on computers can recognise some languages on the device itself; where yours can’t, the page says so, and the browser’s online speech service is used only if you tick “Allow my browser’s online speech service” (the audio then goes to Google, Apple or Microsoft, whichever made your browser). It listens to the microphone only.
• An on-device model cuts the sound at the speaker’s pauses and recognises each phrase on your device: Fast English (Moonshine tiny), Fast (Whisper tiny, every language) or Better (Whisper base). It works in every modern browser and for a tab’s sound, a second or two behind the speaker. On a slow device it can fall behind; the page says so and skips phrases to catch up.

Captions you can read

Four sizes up to extra large, two, three or five lines, the page’s colours or high contrast (white or yellow on black), and full screen. The screen is kept awake while captions run, where the browser allows it.

Translation

Each finished line can be translated as it appears: by your browser’s built-in translator where it has one (Chrome and Edge on computers, for many languages), or by an on-device model for English ↔ Hindi (downloaded once after you agree). Nothing is sent to MySmartCoPilot. Machine translation of live speech makes mistakes; read it as a help, not as the words said.

Limitations

• These are machine captions: they can be wrong and lag behind. They are not a certified captioning (CART) service for legal, medical or exam settings.
• Recognition is best with one speaker close to the microphone; crosstalk, music, distance and accents cause mistakes. Speakers are not told apart.
• A tab’s sound can be captured only in Chrome and Edge on a computer. Phones caption the microphone.
• Your browser’s online speech service sends the audio to its maker; it is off unless you switch it on.
• The free preview captions five minutes at a time and shows the first half of the transcript; a pass removes both limits.
• Caption other people only with their knowledge and consent.

Privacy

Your browser’s recognition runs on this device where the browser can do that; only if you allow its online speech service does your browser send the audio to its maker (Google, Apple or Microsoft). The on-device models are downloaded from MySmartCoPilot once, after you agree (the size is shown first), and then run on your device. MySmartCoPilot never receives your audio or your captions.