How often should you add timestamps to a transcript?

For a long interview, add a timestamp about every five minutes. Oregon State University Libraries tells transcribers to "Add timestamps at five-minute intervals" ¹. For a meeting, add a timestamp at each change of speaker, since that is where a reader most needs to know who spoke. Use word-level timing only when the stamp has to land on a particular word. Our guide to timestamps and speaker labels explains each kind of timestamp.

Five minutes is one archive's convention, not a law

The five-minute interval comes from one archive's oral history guide, where a reader wants to jump to a place in a long recording. It is a sensible default for that kind of work, not a standard that applies everywhere. A one-hour recording given a stamp every five minutes gets 12 timestamps ². That is enough to move around a long file, but it says nothing about the words between the marks.

Short recordings need less. A five-minute interview may not need any interval at all, if you only need the start and the end. The question is what the reader will do with the transcript. If they will search it for a name or a phrase and then listen, the interval should be short enough to find the spot quickly.

Speaker changes are a better trigger for meetings

In a meeting, the most useful place for a timestamp is where a new person begins to speak. Teams live transcription shows this pattern: it "includes each speaker's name and time stamp" ³, so each entry carries the speaker's name with its time. That gives a reader the identity and the time together.

A fixed interval would miss the turn that matters most, because a decision can be made in the middle of a five-minute block. A stamp at each speaker change puts the marker where the conversation changed hands. This is a judgment about what a reader needs, so choose it on purpose rather than by default.

Exact timing needs a different setting

Some uses need timing to the word, not the sentence. Captions and clips are the common cases. Whisper's own timestamps are "at the utterance-level, not per word, and can be inaccurate by several seconds" ⁴, so a sentence stamp is not precise enough for a caption that must change on a particular word.

Precision also depends on the tool. A 2024 study of a fine-tuned Whisper model found that adjusting its tokenizer "significantly improves the precision of word-level timestamps" ⁵. The point is that word-level precision is something you configure and check, not something you get for free.

Choose the interval by what the reader will do next

1. Long interviews: a timestamp about every five minutes, plus one at each new topic if the interview has clear sections.
2. Meetings: a timestamp at each speaker change.
3. Captions, clips and audio-linked editing: word-level timing, checked against the audio before export.
4. Any transcript you will quote: at least one timestamp in each paragraph, so the quote can be found in the recording.

Our overview of transcription covers the whole subject. For how to find a passage quickly in a long file, see how to transcribe an interview, and for the labels on a multi-person recording, see how to transcribe audio with multiple speakers. For the free tools that produce these stamps, see how to transcribe audio with timestamps for free.