Verbatim Transcription

What is verbatim transcription?

Verbatim transcription converts speech to text exactly as spoken. It preserves filler words (um, uh, you know), false starts, repetitions, stutters, and non-verbal sounds such as laughter and pauses. Nothing is paraphrased, corrected, or cleaned up. The finished transcript records what was said and how it was said.

That second part is the point. Most transcription quietly tidies speech: it drops the stumbles, closes the gaps and hands you something readable. Verbatim transcription refuses to tidy. A witness who hesitates before answering, a research participant who restarts a sentence three times, a source who laughs at their own claim: all of it stays in the text.

The word comes from Latin, meaning word for word, and that is still the working standard. If it was uttered, it appears in the transcript. Court reporters have held to this discipline for more than a century, and qualitative researchers adopted it because in interview analysis, hesitations and self-corrections are data rather than noise.

Verbatim is a spectrum rather than a single format, though. Before you order or produce a transcript, you need to know which of the three recognized styles the job calls for.

Full verbatim vs. intelligent verbatim vs. edited transcription

Transcription styles differ by how much of the raw speech survives on the page, and choosing the wrong one can cost you a usable record or bury your reader in disfluencies.

Full verbatim (true verbatim)

Everything is transcribed. Fillers, false starts, repetitions, stutters, interruptions and crosstalk all stay, and non-verbal cues are noted in brackets: [laughs], [pause], [phone rings]. Also called strict verbatim, this is the standard for testimony records, discourse analysis and any transcript that could face scrutiny later.

Intelligent verbatim (clean verbatim)

Also called non-verbatim transcription. Every meaningful word is kept, but fillers, false starts and repetitions are stripped out, and casual forms like "wanna" are standardized. The meaning is never altered. Clean verbatim is the default for business meetings, published interviews and podcasts.

Edited transcription

The most polished level. Grammar is corrected, fragments become full sentences and phrasing is smoothed for print. Edited transcripts read like written prose, which makes them ideal for published Q&As and executive summaries, and unsuitable as a record of exactly what was said.

When you need true verbatim (and when you don't)

The style should follow the stakes: choose full verbatim when the transcript itself is the evidence.

Courts, depositions and testimony

Legal proceedings run on complete records. A deposition transcript that silently drops a hesitation or a false start is a different record of the testimony, and opposing counsel can challenge it. Verbatim transcripts also preserve impeachment material: the exact phrasing a witness used the first time. Legal teams use Sonix to turn depositions, hearings, bodycam footage and client interviews into searchable, timestamped transcripts. See how it works for legal teams.

Qualitative research interviews and discourse analysis

In conversation analysis and grounded-theory work, how a participant says something carries meaning: pauses signal sensitive topics, restarts signal reformulated thoughts. Coding from a cleaned transcript throws that signal away, which is why IRB protocols and methods sections so often specify verbatim transcription. Sonix is built for researchers, with speaker labels you can rename to participant codes and exports that drop into NVivo, ATLAS.ti and MAXQDA. Interview workflows are changing fast, and our interview transcription research shows the data.

Linguistic research

Disfluency studies, sociolinguistics and language acquisition research need every token: the stutters, the repairs, the exact colloquial forms. Standardizing "wanna" to "want to" destroys the observation itself.

Media fact-checking and journalism

When a public figure disputes a quote, the verbatim transcript settles it. Fact-checkers need the full sentence with its false starts and qualifiers, because context is usually where the dispute lives.

When clean verbatim is the better choice

If the transcript is a tool rather than a record, choose intelligent verbatim. Meeting notes, content repurposing and internal summaries all read faster without the ums. For recurring meeting transcription, clean and readable beats strict and complete.

How verbatim transcription works in Sonix

Sonix transcribes speech as it hears it, and one setting decides whether the filler words stay in the transcript.

Sonix's automated transcription engine gives you a timestamped, speaker-labeled first draft in minutes. From there, every step is built for verbatim precision.

Filler words, on or off

The Filler words option in your transcription settings controls whether um, ah and hmm are transcribed in English transcripts. Turn it on for verbatim work, or leave it off for a clean verbatim draft. The setting applies to transcripts made after you change it, so set it before you upload a verbatim job.

Word-level timestamps

Every word is timestamped, so you can click any point in the transcript and hear that exact moment in the audio. For strict verbatim work, this is the difference between a spot check and re-listening to the whole file.

An editor made for verbatim review

For strict verbatim requirements, use the in-browser editor to add non-verbal notations such as [laughs] and [pause], plus any false start or filler the AI smoothed over. The audio plays in sync with the text, so a verbatim pass is fast.

Speaker identification

Speakers are detected and labeled automatically, and you can rename labels to deponent names or participant codes that carry through the transcript. Our speaker diarization guide explains how it works.

Exports for legal and research use

Export Word, PDF and plain text with timestamps and speaker labels intact, plus SRT and VTT for captioned video review. There are 18+ export formats in total.

Scale when you need it

Batch-upload multiple files, or automate high-volume verbatim workflows through the Sonix API. Transcription is available in 54+ languages.

Why accuracy matters more in verbatim work

In a verbatim transcript there is nowhere for an error to hide: every word is the record, so every word has to be right.

In a clean transcript, a small transcription error can hide behind the paraphrase. That is why the base accuracy of the engine matters so much. Sonix delivers 99% accuracy on clear audio, which means a one-hour recording needs correction on a fraction of its words rather than a full re-type.

Audio quality still matters: clear recordings with minimal crosstalk produce the strongest first drafts, so careful mic placement in depositions and interviews pays for itself.

What verbatim transcription costs

With Sonix, verbatim costs the same as any other transcription style: you pay by the audio hour, starting at $10 per hour, not by how much speech you keep.

Human verbatim transcription is the expensive end of an expensive service. Professional human transcription typically runs $100 to $150 per audio hour and takes days to turn around, and difficult audio (multiple speakers, background noise, heavy crosstalk) can add 25 to 50% to the bill. Because full verbatim means capturing every disfluency, services typically price it above their standard rate.

The AI does not charge extra for leaving the ums in. For a full breakdown of AI versus human pricing, see our guide to how much transcription costs, or view Sonix pricing.