From Local Captioning to Voiceover: Whisper‑Style Transcription + Kokoro TTS

Know What Each Tool Does

A Whisper-style transcriber turns recorded speech into draft text and timestamps; Kokoro Web turns a written script into synthesized speech. They are separate steps, not one automatic conversion. Review the transcription app's settings to confirm that it really runs on device. Kokoro Web currently focuses on English voices, so use a voice designed for the target language when the narration must be spoken in another language.

1) Transcribe Locally

• Choose a local transcription tool and save its draft text with word or segment timestamps, such as an SRT or VTT file.
• Replay uncertain spans, names, numbers, and overlapping speakers instead of accepting low-confidence words blindly.
• Keep the original transcript as a reference before making editorial changes.

Automatic speech recognition can miss a negation, normalize a number incorrectly, or assign words to the wrong speaker. Correct these errors against the source recording before producing captions or narration. If captions must preserve what was actually said, do not silently rewrite the transcript to make it sound smoother.

2) Edit and Segment

Keep two text versions when the deliverables have different jobs. Captions should follow the recorded speech and keep their original timecodes. A voiceover script may be shortened, clarified, or reordered, but those edits change timing and must be checked against the picture.

Fix names and technical terms, then split narration into short sense units—one sentence or a couple of connected sentences is a practical starting point, not a strict duration limit. Preserve a stable segment ID so you can replace one line without regenerating the full recording. Add punctuation to show pauses, but avoid inserting stage directions unless the voice model is known to interpret them as intended.

3) Generate with Kokoro

• Preview a representative paragraph before processing the full script; check names, abbreviations, and sentence endings.
• Generate and export each narration segment, then use sequential IDs such as 001 and 002 to preserve order.
• Keep a simple log of the approved text, selected voice, speed, and exported filename.

When a line sounds wrong, change one thing at a time. Try spelling out an acronym or rewriting a difficult phrase before changing speed. This makes it easier to tell whether the wording or the voice setting caused the improvement.

4) Assemble and Master

• Place segments in order in an audio or video editor and adjust pauses to fit the visual cuts.
• Check loudness and true peaks against the delivery platform's specification; avoid clipping during normalization.
• Keep a high-quality master and export the final format required by the destination.

Listen once while following the video and once without looking at it. The first pass catches mismatched names, missing lines, and late or early entrances; the second checks whether the narration remains understandable on its own. If you revise the voiceover wording, update the segment log and recheck any captions that are meant to match the new spoken track.

Benefits

• When both tools are configured for on-device processing, source media and script text stay out of processing-server uploads.
• Consistent narration quality and repeatable edits.
• Faster iteration on phrasing and timing.

Frequently Asked Questions

Do I need to upload media?

Not if your transcription app and TTS workflow both run on device. Verify the transcription tool's processing mode; Kokoro Web generates speech in the browser, although model files may need to download first.

How do I sync captions and voiceover?

Keep the original transcript timestamps for captions, create separate narration segments, and realign the audio in your editor after rewriting the script.

What export formats should I use?

Keep a high-quality master and export the format required by your destination. Check timing, loudness, and clipping after export.

Author: Kokoro Web Team • Last updated 2026‑10‑15