Start by defining what must be exact
Not every line in an interview needs the same level of review. If you are looking for themes, the first pass may be enough to identify useful sections. If you plan to publish a direct quote, the exact wording, punctuation, speaker, and surrounding context matter much more.
Make a short review list before opening the transcript: names, organizations, product terms, dates, numbers, claims, and sentences you may quote. This prevents a long recording from turning into an unfocused proofreading task.
Use timestamps as evidence links
The most useful feature of a timestamped transcript is the path back to the source. When a sentence looks unusual or important, click its segment timestamp or an estimated word position and listen to the recording. Check the beginning and end of a quote, not only the single word that looks wrong.
Timestamps are approximate model output, so they should guide review rather than replace it. Pauses, crosstalk, and unclear audio can affect the exact boundary. For publication, confirm the quote in the player and keep the source file available until the edit is complete.
- Read the transcript once to mark names, numbers, and possible quotes.
- Replay each marked segment from its timestamp and compare the wording with the audio.
- Edit the segment only after you understand whether the issue is recognition, punctuation, or editorial style.
Assign speaker labels manually
Interview transcripts are easier to scan when questions and answers have consistent labels. The current local editor lets you assign Speaker 1 through Speaker 4 to each segment. Start with the clearest opening question and answer, then continue through the recording while checking transitions.
These labels are a review aid, not automatic diarization. If the interview has interruptions, two people speaking at once, or a distant microphone, listen before assigning a label. When a segment is genuinely ambiguous, keep it unassigned or mark it for a second review instead of inventing certainty.
Keep a small quote and fact log
For a research or publishing workflow, keep a short list outside the transcript with the timestamp, the proposed quote, the speaker, and the review status. This turns a long interview into a set of auditable decisions. It also helps another editor understand which lines were checked against the recording and which are still only leads.
A quote log does not need to duplicate the entire transcript. Record only the sentences you may use, the facts that need confirmation, and any context that changes their meaning. Link each note to the transcript timestamp so the source can be reopened without searching from the beginning.
If a quote is shortened, note the omission and confirm that the remaining words preserve the speaker's meaning. The transcript editor can hold the readable draft, while the log records the editorial reasoning around the final selection.
Handle uncertain words openly
Some audio remains ambiguous after several listens. Instead of guessing, mark the word for follow-up, ask the speaker or producer when possible, or leave a visible uncertainty note in the working copy. A transparent unknown is safer than a confident but incorrect name or claim.
The browser editor can highlight the segment for another review and keep the timestamp attached. If the local model does not return token confidence for a word, do not invent a numerical certainty score. Use the audio, glossary checks, and human review decisions as the evidence you actually have.
Once the important uncertainties are resolved, export a clean copy and keep the reviewed version separate from the unedited first pass. This makes later corrections easier to trace without confusing a draft with approved publication text.
Check names and terminology
Names and specialist words often look plausible even when they are wrong. Before transcription, add known terms to the names and terms field. Afterward, search the transcript for each one and listen to the relevant audio. The glossary review message can point out terms that were not found in a segment, but it cannot decide the correct spelling on its own.
Use the interview context as a second source of evidence. A company name may be visible in your research notes, while the audio confirms how the guest pronounces it. Resolve the conflict consciously and keep a note of any editorial spelling decision that differs from the spoken form.
Separate transcription edits from editorial edits
A transcription edit corrects what was heard: a missed word, wrong name, or punctuation that obscures the sentence. An editorial edit changes how the material will be presented: removing a repeated phrase, tightening a response, or applying a publication style. Keeping these decisions separate makes the transcript easier to audit.
The editor provides a textarea for each segment so you can make a readable draft without losing its timestamp. For a significant quote, preserve the source wording in your working copy and make any shortening or cleanup visible to the person approving the final text.
Export a copy that preserves your review
TXT is useful for a writer reading the interview, JSON is useful when a developer needs segments and timing, and SRT or VTT are useful when the conversation will accompany video. Export only after the important segments have been checked, because an export is a snapshot rather than a live link to your review session.
For the complete workflow, use the interview transcription tool. If the recording is a meeting rather than an interview, the meeting transcription page explains how to handle decisions, action items, and manual participant labels.