Recipe videos. Keep every measure clear.

Subtitle the demonstration you actually recorded

A cooking video can say one quantity while an ingredient card shows another. A missing fraction, an unfamiliar ingredient or a translated unit can change what viewers understand. Polyglot creates a timed subtitle draft from your narrated recording on an Apple Silicon Mac. Review the source, optionally translate it and export a separate SRT or WebVTT file.

Use footage you own or have permission to process and distribute, including translated publication. Check the use of collaborators’ appearances, music and visible third-party material. Remove uncleared footage and personal details before importing the final edit. This workflow prepares subtitles; it does not validate a recipe or certify food-safety instructions.

Finish cuts, speed changes, voiceover and ingredient overlays first. Keep the matching recipe revision beside the video. A silent overhead montage has no narration for speech recognition to transcribe; its visible instructions need their own authored text or description.

Compare the speech, the card and the recipe note

Import the finished video, or an existing SRT or WebVTT that matches it. Listen while reviewing the draft. Keep a separate discrepancy log with the time, exact spoken words, visible card and question for the recipe owner. The log is a manual review aid, not an automatic comparison feature in Polyglot.

Distinguish a transcription mistake from a recording mistake. If the presenter says “teaspoon” but the draft says “tablespoon,” correct the draft to match the speech. If the presenter and ingredient card disagree, do not silently choose whichever looks plausible. Ask the owner to resolve the source conflict with a corrected recording or clearly identified editorial correction. Then review the matching subtitles again.

This fictional review note illustrates a publication hold. It is not a recipe, an app-generated report or evidence of transcription accuracy. The amount is deliberately unresolved until the creator checks the source.

Fictional review note
Video: recipe04-v03.mp4
00:18 spoken: “one teaspoon”
Card: “1 tablespoon”
Draft: “one teaspoon”
Hold: owner resolves mismatch
Then recheck video + tracks

Keep quantities attached to their ingredients

Make a dedicated pass for fractions, decimals, quantities, units and ingredient names. Replay unclear speech instead of reconstructing a likely recipe. Keep qualifiers such as “divided,” “optional,” “room temperature” or “finely chopped” when spoken. Check which ingredient a pronoun refers to when the shot changes.

Before translating, ask the creator for approved ingredient names and any regional terminology. Keep a short reference sheet for the reviewer; this is not a glossary-import feature. An unfamiliar ingredient should not quietly become a substitute. Preserve instructions such as “add gradually” and negations such as “do not boil,” rather than making them more definite or convenient.

Keep the original quantities and units during the language pass. Translation is not permission to scale portions or convert cup measurements to grams. King Arthur Baking’s ingredient chart gives different weights for different ingredients and measures, illustrating why a generic cup conversion is insufficient. Any localized measurement alternative needs separate creator review against the actual recipe.

Compare the translated track with both the reviewed source and the final video. A decimal separator or fraction must still mean the same amount. If the presenter did not specify a temperature unit, flag that omission for the owner; do not guess it from the audience’s country. Preserve the distinction between a cooking duration and an instruction to wait for a visible change.

Review silent steps and meaningful kitchen sounds

Listen for sounds that carry information: a timer ringing or a change in sizzling may matter to the explanation. W3C’s caption guidance includes meaningful non-speech sound and speaker context as well as dialogue. Polyglot does not automatically produce a complete sound-description track or identify every speaker. Where a sound needs a new timed cue, use a caption editor that can author and time it.

Watch once without sound and note essential instructions that appear only visually. Polyglot’s audio transcription does not read ingredient overlays, scales or oven displays. Captions for speech do not automatically describe a silent folding technique or a texture change. W3C’s guidance on visual description treats essential visual information as a separate planning task; consider narration or another appropriate description when that information is needed.

Keep subtitles faithful to the audio and make additions clearly identifiable. Do not insert an unseen ingredient into a speech cue as though the presenter named it. Resolve critical gaps in the video or supporting material with the creator before release.

Check the subtitles while the hands are working

Watch the whole finished video with each reviewed language track. Start by checking the first spoken instruction, a distinctive middle step and the final line for synchronization. Then inspect quantities, ingredient qualifiers, cue order and any unexplained missing text. Translations retain source timing, so longer wording still needs a reading-time review.

Use the destination player at a small screen size. Keep the captions readable without covering a scale reading, ingredient overlay or the hand movement being explained. Review lines after a pause and seek, and confirm that only the intended caption track is showing. A sidecar file cannot guarantee identical placement or styling in every player.

For YouTube, its documented workflow lets you select a language and upload a file with timing. Check the destination’s accepted formats and review the uploaded result. Polyglot exports separate SRT or WebVTT files; uploading and any required burn-in happen in other tools. A shortened recipe clip needs subtitles matched to its own edit.

Keep the final video, approved recipe note, resolved discrepancy log and reviewed language files together. Polyglot processes media and subtitle text locally. Requested downloads, user-controlled updates, default-on content-free analytics with an opt-out and Pro licensing have separate network boundaries. New AI subtitle jobs use the seven-day trial and require Pro afterward; editing and exports remain free.

Keep exploring.