Why this step matters most
Everything a learner sees is built from the transcript: the lines they read, the words they collect, the translations, the difficulty, the questions. A wrong transcript does not produce a slightly worse video — it produces a video that teaches the wrong thing. So this is the one step Roxxem will not guess at for you when it has any doubt.
Open step 2 — Transcript on a video's setup page.
Compare the candidates
Roxxem gathers every transcript it can find and lists them as candidates. Each one shows where it came from and how it was graded, and you can read its lines before choosing.
What to look for:
- Does it stop early? A transcript that covers the first minute of a four-minute video is a failed transcription, not a short one.
- Does it repeat? Loops of the same phrase are a known failure of machine transcription.
- Do the timings move sensibly? Lines that all start at the same second, or that run past the end of the video, are broken.
- Is it the right language? A transcript in the wrong language happens more often than you would think.
- For a song: are these the actual lyrics? Machine transcripts of singing are frequently wrong in ways that read fluently. Compare against the lyrics if you know them.
When one looks right, click Select. Roxxem then asks you to confirm, and shows you what selecting it will do.
Or paste your own
If no candidate is good enough — common for music, and for anything with heavy background noise — use Manual transcript entry and paste the lines yourself. One line per line. Timing comes later, on the edit page.
Hand-copying lyrics or a script from elsewhere has its own rules, and they matter legally as well as pedagogically. Read the guidelines on where the text comes from before you paste.
How lines should be broken
A transcript is not just the right words, it is the right words in the right pieces. Line length, where a line should break, what to do with a sentence that runs across two shots — all of it is in the guidelines on lines and on timing.
What happens next
Once a transcript is settled, the video continues to step 3 — Generate on its own. You do not have to trigger anything.
If you later decide the transcript was the wrong choice, you can come back to this step and select a different candidate — but be aware that changing the transcript replaces everything built from it. How to re-run a video through the pipeline explains what survives and what does not.
Songs are held back on purpose
A song never publishes off an unchecked machine transcript. If a song is sitting at step 2 or step 4 waiting for you, that is the system working: someone has to read the lyrics. The same is true of languages where our line-breaking is not yet trustworthy — see the language-specific rules.