Skip to content
100% local

Speaker label normalizer

Clean up speaker labels in a transcript, rename them and see how much each person talked.

Input
Name mapping (optional)
Output

Speaker label normalizer

Paste a transcript from a meeting recorder, podcast editor or captioning tool and this tool cleans up how the speakers are labeled. It recognizes lines like "Speaker 1: text", "SPEAKER_00: text" or "Jane: text", rewrites every label in one consistent style, and tells you how many words each person spoke.

Choose a name style to control how the cleaned-up label looks — "Name:", "[Name]:", "Name -" or "NAME:". If the source only has generic tags such as Speaker 1 or Speaker 2, list the real names in the mapping panel, one pair per line as "Speaker 1=Jane Doe", and every matching label is replaced. Turn on "Merge consecutive turns" to combine several lines from the same speaker in a row into one paragraph, and "Remove timestamps" to strip bracketed time codes some tools insert next to a name. Switch the output to "Speaker statistics" to get a plain word-count and percentage breakdown per speaker instead of the full transcript, or keep the transcript and add that same breakdown underneath it with "Add word count and speech share per speaker."

The tool handles Windows (CRLF) and Unix (LF) line endings, keeps continuation lines that were split across multiple rows attached to the right speaker, and leaves any line it can't confidently read as a label untouched rather than guessing. The live tally below the output shows lines in versus out, how many were folded together by merging, and the total word count.

Everything runs locally in your browser — a transcript never leaves your device. When you're happy with the result, copy it, download it as a .txt file, or send it straight into another tool like the voice transcript cleaner or the subtitle generator to continue editing.

FAQ

What speaker-label formats does it recognize?
Lines that start with a name followed by a colon, such as "Speaker 1: text", "SPEAKER_00: text" or "Jane Doe: text", with an optional timestamp in brackets or parentheses before or after the name.
How does the name mapping work?
List one pair per line in the second panel as OldName=NewName, for example "Speaker 1=Jane Doe". Matching is case-insensitive, and any label without a matching row is left as it appears in the transcript.
What happens to lines that don’t look like a speaker label?
A line with no recognizable "Name:" pattern is treated as a continuation of the previous speaker’s turn, or kept as plain text if it appears before any label.
How is the speech-share percentage calculated?
It divides each speaker’s total word count by the total words spoken by all labeled speakers in the transcript, so it reflects the whole conversation regardless of whether turns are merged.
Is my transcript uploaded anywhere?
No. Speaker label normalization runs entirely in your browser — your transcript never leaves your device.