Edit video by editing text: a solo streamer's first transcript-editing session

Published November 14, 2026 · Workflow test + official plan summary

If you publish alone, editing is probably where your week goes. Transcript-based editing attacks the slowest part first: instead of scrubbing a timeline for every "um" and every sentence you said twice, you read the transcript, delete the words, and the video cuts itself to match.

The test recording

Our sample was a short "weekly planning" monologue for a streamer. We deliberately wrote in the mistakes real recordings have:

• Four fillers: "Um…" twice, "Uh…" once, and one "you know"
• One sentence said twice in a row: "I build every session around that one game."

The recording was synthetic speech generated from our own script (we say so openly: results on a synthetic voice do not prove accuracy on your microphone, your room, or your accent). These are examples of issues to check for, not predictions of what will happen on your own recording.

The workflow, step by step

1. Record or import. One file in: our sample was a 40-second 720p video.
2. Transcribe with word-level timestamps. Our transcription pass over the 39.5-second audio took about 5 seconds with a local speech-to-text model, and it found all four fillers and both copies of the repeated sentence.
3. Read the transcript and mark the cuts. This is the whole point: finding a bad sentence in text takes a glance. Finding it on a timeline takes a listen.
4. Cut, with padding. On our first attempt we cut exactly the timestamps the transcript gave us — and the re-check still found filler fragments ("Um" and an "Oh" where "Uh" had been). In this sample, some word timestamps did not cover the full audible filler. Cutting about 0.2 seconds wider on each side fixed it.
5. Re-transcribe to verify. After the second edit, a fresh transcription of the result found zero fillers and exactly one copy of the repeated sentence. A second transcription can flag possible mistakes, but it can miss errors too. Listen to the complete output before publishing.
6. Export the video and the captions from the same pass. The edited transcript is the caption file. We exported an SRT with six caption blocks for the edited video.

The numbers from our test

The handoff log reports 39.5 seconds of source audio, a 40.9-second source video, and a 34.2-second edited output. These different baselines do not support one precise video-reduction percentage. Original media was not included in this handoff, so the figures have not been independently reproduced. No editing-time saving on long recordings was measured.

Where Descript fits

Descript is one tool that supports this workflow: you record or import, it transcribes, and you edit the video by editing the text — including one-click filler-word removal and captions. That is why it is the main tool this site covers for editing.

What the free plan includes, per Descript's official pricing page (checked November 14, 2026):

• Price: $0. The pricing page's FAQ answers "No" to needing a credit card to try Descript, and the sign-up screen asks for Google, single sign-on, or email — no card field.
• Media time: 1 hour per month on the free plan.
• Export: 720p.
• Paid plans (for when the free hour runs out): Hobbyist at $16/month billed annually ($20 monthly), Creator at $24/$30 at the time we checked.

Two honest caveats. First, Descript's own pages disagree in places — its tool pages describe the free export as watermark-free while the pricing page's feature table lists watermark-free export as not included on Free — so check the pricing page before you rely on either claim. Second, 1 media hour a month is a trial allowance, not a production allowance: it is enough to learn the workflow on your own clips and decide, which is exactly what a free plan should be for.

Try Descript's free plan (official site)
Plain official link — not an affiliate link yet. See the disclosure at the top of this page.

Who this workflow is for — and who should skip it

• Good fit: talking-head videos, podcasts, stream recaps, tutorials — anything where the words carry the content and most cuts are verbal mistakes.
• Poor fit: gameplay-first or music-first edits where the picture leads and the transcript is thin. A timeline editor will serve you better there.
• Do the verify pass: whatever tool you use, re-transcribe (or at least re-listen to) the edited result once. Our first cut sounded fine on paper and still had filler tails in the audio.

Plan facts change. Descript's pricing page is the final word; this article was last checked against it on November 14, 2026.

Test provenance: the previous project agent reported running Piper TTS (en_US-amy-medium), faster-whisper base on CPU int8, and ffmpeg on November 14, 2026. This article describes that local-tool test, not a session inside Descript. The current handoff contains the written log but not the original audio, scripts, or generated media.

Sources and review notes

• Descript pricing

Editorial review: November 14, 2026. Unsupported interview and time-saving claims were removed; test limitations and source discrepancies are stated explicitly.