Check Interview Names and Quotes

The short answer

Keep transcription in the spoken language, then click the time beside a passage to replay it with the surrounding question. Correct confirmed words in the text. Record uncertain words separately, verify names and figures, and export a version with time references for the next review.

Open the free tool

Keep the original wording separate from translation

For an interview that you plan to quote in its original language, leave “Translate into English while transcribing” off. Set “Spoken language” when you know it, or review what automatic detection produced. A translated sentence is useful for understanding, but it is a different editorial product from the words spoken. If an English translation is also needed, preserve the original-language passage and identify the translation separately in your working document. OpenAI’s Whisper model card, checked on 22 October 2026, notes that recognition can produce words absent from the audio. A polished sentence therefore deserves listening, not extra confidence simply because its grammar looks convincing.

Use the timestamp as an entrance to the conversation

Click the time next to a relevant segment to start playback there. Listen beyond the sentence you intend to quote and return slightly earlier when you need the question. An answer such as a short agreement can mean very little without knowing what the interviewer asked. In an illustrative newsletter interview, an owner’s statement about opening on Sundays might refer to a temporary event rather than normal hours. Keep that context with the proposed quotation in your notes. The timestamp locates a segment, not a certified boundary for the quote. Verify where the actual words begin and end in the recording instead of copying every word of the displayed segment automatically.

Build a spelling list from identifiable information

Keep a separate list of names, places, organisations and specialist terms mentioned in the interview. Use spellings supplied by the speaker or another identifiable original source, and note where each came from. A recogniser may turn an unfamiliar surname into an ordinary word; correcting that is different from guessing which similarly named person was meant. Click the sentence to edit its text once the spelling is established. Check later occurrences too, including abbreviated forms. Do not assume an unfamiliar term is wrong only because a browser spellchecker underlines it. When two spellings remain plausible, preserve the question for clarification rather than selecting whichever looks more familiar.

Listen closely to numbers and small qualifying words

Review amounts, dates, quantities and units in every passage you expect to use. In a fictional example, a discussion of fifteen deliveries is materially different from fifty, even though both fit the topic. Write down the unit with the number so an hourly figure does not become a daily figure in the article. Pay particular attention to a missing negative, a conditional phrase or a qualification such as approximately. Those words may change the meaning more than an obvious spelling error. Do not use a calculation to manufacture what the speaker must have said. If the recording and a supplied document disagree, keep both observations visible and resolve the discrepancy as an editorial question.

Attribute turns by listening, not by paragraph shape

Transcript Studio presents timed segments without automatic speaker names. A new segment is therefore not proof that a different person started speaking. Listen for each change of voice, especially when an interviewer interjects or two people overlap. Add speaker labels manually only when the recording supports them, either in the edited text or in your exported working copy. Decide on a consistent convention for the interviewer and guest. If an overlapping phrase cannot be assigned confidently, mark that uncertainty instead of moving it into the answer you prefer. This matters when several people use similar expressions or a short acknowledgement sits between two longer answers.

Leave unresolved audio unresolved

For a difficult word, replay the passage with the preceding and following speech. A slower playback setting can help you listen, but it cannot recover a sound that was never captured clearly. Keep an explicit uncertainty note in your working document, with the time and the competing interpretations if useful. Make clear that the note is yours, not speech recognised from the recording. Ask the interviewee for clarification when that fits the editorial process, without treating the answer as proof of what was audible in the original. Repeated words and a fluent sentence over a silent passage deserve particular attention. Never fill a gap merely to make the transcript read smoothly.

Distinguish a corrected transcript from an edited quotation

Correcting an incorrectly recognised word aims to recover the recording. Removing hesitations or combining parts of an answer is a later editorial decision that should remain visible in your working process. Keep a source-faithful version before creating a shortened article draft. Read the proposed quotation back against the audio after cutting it, because nearby context may limit what the sentence claims. Do not turn several separated replies into one uninterrupted quotation without acknowledging the edit according to your publication’s practice. This guide does not set a universal quotation policy. It gives you a way to preserve the evidence needed by the person making that decision.

Search for another occurrence, then listen again

Use “Search in the text” to find repeated names or a phrase you have just corrected. The search checks the recognised text, so a missing result does not prove that the recording never contains the word. Try a shorter distinctive part or a plausible recognition variant when necessary. Use the matches to revisit the relevant segments and check them by ear. Search is especially helpful when a business name first appears in full and later in a shortened form. Avoid replacing every similar string blindly: the same letters may occur in an unrelated word, or the speaker may genuinely use two different terms. Each changed meaning still needs a contextual check.

Export a review copy that keeps the route back to audio

Choose “Text with times” for a plain working transcript with segment references, or “Word (.docx)” when the reviewer wants an editable document. Keep the recording available under the same version reference so the times lead to the correct interview edit. The export carries your current text corrections; it is not an independent approval of them. In a separate review note, identify unresolved names, disputed numbers and the exact proposed quotes awaiting a decision. Do not rely on JSON as a promised project-reopening feature: the current tool exports segment data, while review can continue in the saved document. Before using “New file”, save the version you need and check that the chosen export contains the corrections.

Questions you might still have

Does the tool identify the interview speakers?

No. It supplies timed segments without automatic speaker names. Attribute turns manually after listening.

Can I edit a recognised sentence?

Yes. Click its text and correct it. The edit changes the transcript, not the original recording.

Which download retains playback references?

“Text with times” includes segment references. Word export also includes times in the current implementation.

Can I quote an English translation as original wording?

Keep the translation distinct from the original-language transcript and review it as a translation. The English option changes the output language.

Does search find every spoken occurrence?

It searches the recognised text. A misrecognised or omitted word may need a different query and manual listening.

What should I do with an inaudible name?

Keep a dated review note with the time and seek clarification. Do not turn a plausible guess into a confirmed spelling.

Sources & useful details

These guides are written from the current LoschaZ tools and the sources listed above. Examples are illustrations, not tests on every device or promises of an identical result.