Standard or Enhanced speech? Clear voice stays fast | AI Video Factory

Slow transcripts and mangled names are often the recognition mode, not the quota. The studio has two: Standard (faster) and Enhanced (more accurate). Clear voiceover starts on Standard. Heavy accent, dense names, or a noisy room moves to Enhanced. Pick before you run. After you close the page, the free plan cannot re-listen in a different mode this month.

Confirm you are transcribing

Recognition mode only affects speech. Burned-in type and almost no voice belong to on-screen OCR; switching mode will not read the picture. How to choose: speech or OCR. Cues that already sit on the mouth: import them — skip recognition.

Standard is enough for one clear speaker, a quiet room, and few proper names. Enhanced is worth the wait for mixed accents, brands and model numbers, or noise where the voice is still audible. If you cannot hear the words, cut the empty heads first. Do not hope a mode change fixes silence.

Pick once before you start

1. Open captions in the studio and drop a local MP4 / MOV. The file stays in the browser.
2. After detection, tick speech recognition. Under Recognition mode, choose Standard or Enhanced. Language can stay on Auto — ten-minute captions.
3. The first Enhanced run downloads a larger local model. A short, clear clip does not need that wait.
4. Review names and numbers. If the page is still open and you picked wrong, switch mode and listen again. After you close it, this month’s free caption run is spent.
5. Then style or export — vertical layout and SRT or burn-in.

The free plan includes one caption run per month. Modes and limits: Features and Pricing.

Check only three things

• Voice: one clear speaker → Standard. Overlapping talk: cut first, do not jump to Enhanced.
• Names: dense brands, models, and people → Enhanced, then still scan them yourself.
• The run: change mode before you start. After the page closes, wait until next month or upgrade.

What usually breaks

A clear 30-second voiceover on Enhanced: a longer download, almost the same accuracy. Typos after Enhanced, then rerun after rerun: the model is for speed, you own the last pass. Treating Enhanced as OCR: it does not read type on the picture.