Transcription With Timestamps: Word-Level Timing and Speaker Labels

A plain transcript is a wall of text. You can read it, but you cannot do much else with it. You cannot turn it into captions, you cannot jump to the moment someone said a particular sentence, and you cannot tell who was speaking when three people are in the room.

Here is what a transcript looks like when every word carries its own timing. Press play and watch it track the audio, word by word. This is the same sample and the same timestamps you get from the speech to text tool.

That highlight is not an animation timed to the clip. Each word has a real start and end time attached to it, which is what makes captions, search, and editing by text possible.

Timestamps are what turn a transcript into something workable. Once every word carries a start and end time, the same file becomes subtitles, a searchable index, an editing map, and an alignment track for dubbing. Add automatic speaker labels on top and you also get a record of who said what.

This guide covers what the different timestamp granularities actually mean, which one you need for which job, how to produce timestamped transcripts (with speaker labels) in Voice Creator Pro, and which export format to pick.

Timestamp Granularity: Segment, Word, and Character

Not all "transcription with timestamps" means the same thing. There are three levels, and the difference matters more than it sounds.

Granularity	What you get	Best for	Typical availability
Segment / sentence level	A start and end time for each block of text, usually a sentence or a caption line	Basic subtitles, rough navigation	Almost every transcription tool
Word level	A start and end time for every individual word	Karaoke captions, text-based audio editing, search, dubbing alignment, clip extraction	Available in Voice Creator Pro and a handful of others
Character level	Timing for individual characters within a word	Almost nothing in practice, see below	Rare

Segment level is the default almost everywhere, because it is all a basic SRT file needs. Each subtitle line gets an in and out time and that is it. It is fine if all you want is captions that appear and disappear on cue.

Word level is where the transcript becomes a tool rather than a document. When each word carries its own timing, you can highlight words one by one as they are spoken, cut audio by deleting text, find the exact second a phrase occurs, and line a dub up against the original delivery.

Character level comes up in searches, but it is worth being straight about it: speech recognition models predict words and sub-word tokens, not characters, so character timings are interpolated rather than measured. For every real workflow (captions, editing, search, alignment) the word is the meaningful unit. If you were looking for character-level timing, word-level is almost certainly what you actually needed.

Voice Creator Pro produces word-level timestamps, so every word in the transcript carries its own start and end time.

What Word-Level Timestamps Are Actually For

Five jobs make up nearly all of the demand:

1. Captions and subtitles that track the voice. Segment timings give you a block of text on screen. Word timings let you highlight each word as it is spoken, which is the style that holds attention on short-form video.
2. Editing audio by editing text. With per-word timing you can find the exact boundaries of a filler word, a flubbed take, or a whole sentence and cut it out cleanly, without scrubbing a waveform.
3. Making a back catalogue searchable. Index the transcript, and a search hit points at a timecode rather than an episode. This is what turns a podcast archive or a course library into something people can actually navigate.
4. Aligning a dub or a translation. To fit a translated line into the space the original occupied, you need to know how long the original took. Word timings give you that per phrase, not just per paragraph.
5. Pulling clips automatically. Given a quote, word timings tell you exactly where to cut. That is how a long interview becomes a set of social clips without a manual pass.

Speaker Identification: Who Said What

Timestamps tell you when something was said. On multi-speaker audio (interviews, panels, meetings, two-host podcasts) you also need to know who said it, otherwise the transcript is an undifferentiated stream and you have to reconstruct the conversation by ear.

Voice Creator Pro handles this automatically. When you transcribe audio containing more than one voice, it detects how many distinct speakers are present and attributes each part of the transcript to the speaker who said it. You do not tag anyone in advance, and you do not tell it how many people to expect. Those speaker labels carry through into the exported files as well as the on-screen transcript, so whatever you build downstream keeps the attribution.

Practically, that is what makes these usable:

• Interview and podcast transcripts that read as a conversation instead of a monologue.
• Meeting notes where you can see who raised a point and who committed to an action.
• Two-column or coloured captions for videos with more than one person on screen.
• Per-speaker dubbing, where each speaker in a video is re-voiced separately. Voice Creator Pro's video dubbing pipeline uses the same speaker separation to assign a different voice to each person in the source video.

A caveat worth setting expectations on: speaker separation is hardest exactly where you would expect it to be, on heavy crosstalk, on people talking over each other, and on two voices that sound very similar. Clean audio with distinct speakers separates cleanly. A four-way argument recorded on one room mic will need a pass by hand.

How to Transcribe Audio With Timestamps

The steps below are the same whether you run the desktop app or use Voice Creator Pro Cloud in your browser.

Step 1: Open the STT tab

Click Lab in the left sidebar, then select the STT tab. The Audio Source panel is on the left, the Results panel on the right.

Step 2: Load your audio

There are four ways in:

• Upload an audio file from your computer (WAV, MP3, and other common formats).
• Record straight from your microphone.
• YouTube, by pasting a video URL and letting the app extract the audio.
• History, to pull in something you already generated in the Clone, Design, or TTS tabs.

Step 3: Set the language

Pick the spoken language from the Language dropdown, or leave it on Auto-detect. If you know the language, select it explicitly. It gives the model a head start and tends to help on short or noisy clips.

Step 4: Choose an ASR model

Click the model badge in the Results panel to switch between the available ASR (Automatic Speech Recognition) model families. The default is a good general choice. If accuracy is disappointing on a particular recording, switch families and run it again, since different models cope differently with accents, background noise, and fast speech.

Step 5: Transcribe

Click Transcribe. The text appears in the Results panel, with word-level timing captured for every word and each speaker attributed on multi-speaker audio.

Step 6: Export

Export as SRT or JSON from the Results panel. Which one you want depends entirely on what happens next, which is the section below.

For a fuller walkthrough of each workflow, see the transcribe audio tutorial, and the Speech to Text reference for every setting.

SRT or JSON: Picking the Right Export

Format	Contains	Use it when
SRT	Numbered caption lines with in and out times	You are dropping subtitles into a video editor (Premiere Pro, DaVinci Resolve, CapCut, Final Cut) or uploading a caption track
JSON	The full transcript plus word-level timestamps in a structured object	You are writing code against it: custom caption styling, search indexes, clip extraction, alignment, any automated pipeline

The short version: SRT for video editors, JSON for code. SRT is a caption format and flattens timing to the line, so if you need per-word timing you want JSON.

If your end goal is styled captions burned into a video rather than a sidecar file, the subtitle generator is the better route. It builds on the same transcription, adds caption and karaoke-style highlighting, background boxes and fonts, and can translate the caption track into 21 languages, exporting either burned-in captions or SRT/VTT sidecars.

The desktop app also exposes transcription through a local REST API, so you can post an audio file and get the timestamped JSON straight back into your own tooling. That API is desktop only and is not part of the cloud version.

Getting Better Timestamps

Timestamp precision is downstream of transcription accuracy. If a word is misheard, its timing is meaningless. A few things move the needle:

1. Feed it the cleanest audio you have. Use the original recording, not a re-encoded copy or the audio pulled off a compressed upload. Background music and room noise cost you more accuracy than any setting will win back.
2. Set the language explicitly when you know it, rather than leaving auto-detect to work it out.
3. Try a second ASR model before concluding the audio is the problem. Model families differ noticeably on accented and fast speech.
4. Record speakers on separate tracks when you can. Nothing improves speaker attribution like the speakers not overlapping on the same channel.
5. Check the first thirty seconds before processing an hour. Run a short section, confirm the language and model are right, then commit to the full file.

Where to Run It

Transcription with timestamps is available both ways:

• Voice Creator Pro Cloud runs in the browser with nothing to install, and has a free tier of 10,000 tokens a month with no card required. Try it free in your browser.
• The desktop app processes everything locally on your own machine, so no audio leaves your computer and no internet connection is needed. It is a one-time purchase with no subscription, and it is the option to pick for confidential recordings or bulk work. See desktop pricing.

Both produce the same word-level timestamps and speaker labels, and both export SRT and JSON.

Related Reading

• Speech to Text, the product page for the transcription tool
• How to add subtitles and dubbing to YouTube videos
• How to dub a video into another language