Spoken-word transcripts

What was said, turned into
text you can actually read.

Paragraphing, punctuation, speaker labels, and filler removal turn a spoken-word transcript into text you can quote. Nothing is added, facts and figures stay as spoken, and unclear audio is marked [inaudible] rather than guessed.

Nothing addedFigures unchangedDoubt marked

Review the cleanup

Your result will appear here

We add paragraphing, punctuation, and speaker labels, and list the [inaudible] passages you should confirm against the audio.

SAMPLE OUTPUT

Here is exactly what you receive.

This is the complete output for the source text in the demo above: a readability score, itemised cleanup notes, passages to confirm against the audio, and the cleaned transcript. It is a demonstration, not a customer case.

SAMPLE

Cleanup result (sample)

Readability score88/100

+50

Add paragraphing and punctuation

The source was one unpunctuated stream of speech. It becomes two sentences with punctuation, so the reader does not have to parse it alone.

Remove fillers and say how many

“Um”, “so”, “then there is”, and “I guess” are verbal padding. Four were removed; the meaning is unchanged.

Mark the unclear passage [inaudible]

The audio is muddy at that point. Its position is flagged rather than filled in, so no unverifiable quotation enters the record.

Cleaned transcript

# Release fix notes (cleaned transcript)

**Speaker A**: This release mainly fixes the import failure. The root cause is a file encoding problem. After it ships we still need to watch it for about a week.

**Speaker A**: [inaudible] (the audio is unclear here; may concern a specific metric)

**Cleanup notes**: removed 4 fillers ("um", "so", "then there is", "I guess"); added punctuation and paragraph breaks; one speaker throughout, judged from the source.

**Unchanged**: every figure and fact is reproduced as spoken, with no conversion or correction.

**To verify**: replay the unclear passage; if it turns out to be a metric, do not quote it directly.

WHAT CLEANUP DOES

Not rewriting the speaker. Making their words readable.

Structure moves, content does not

Paragraphing, punctuation, filler removal. Any edit that would change meaning is off limits.

Speakers must be traceable

If the source does not say who is speaking, we use a neutral label like Speaker A and flag it, instead of inventing a name.

Mark what you cannot hear

Uncertain dictation is marked [inaudible] so you can replay it. A guessed word becomes a fabricated quotation.

WHY IT IS WORTH DOING

A transcript nobody reads is no record at all.

The raw text is barely readable

Speech arrives as one unpunctuated stream. Readers give up after two lines and the meeting may as well not have happened.

Cleanup tends to change meaning

Smoothing “I am not sure that is true” into “I think it is wrong” flips a position in one move.

Transcripts get used as evidence

Meeting notes, interview quotes, and podcast captions all get revisited. Altered facts and figures carry real risk.

Note: speech differs sharply from writing — fillers, repetition, and self-correction are normal in natural talk, and verbatim, clean, and caption transcripts serve different purposes. This site works at the level of structure and punctuation only. It does not change facts or figures and provides no legal advice. For evidentiary use, rely on the original recording.

CLEANUP DEMO

One passage, two versions.

Before

um so this release mainly fixes the import problem and then there is the encoding thing I guess after it ships we still need to watch it for like a week the specific metric is still being checked

Cleaned up

# Release fix notes (cleaned transcript)

**Speaker A**: This release mainly fixes the import failure. The root cause is a file encoding problem. After it ships we still need to watch it for about a week.

**Speaker A**: [inaudible] (the audio is unclear here; may concern a specific metric)

**Cleanup notes**: removed 4 fillers ("um", "so", "then there is", "I guess"); added punctuation and paragraph breaks; one speaker throughout, judged from the source.

**Unchanged**: every figure and fact is reproduced as spoken, with no conversion or correction.

**To verify**: replay the unclear passage; if it turns out to be a metric, do not quote it directly.

Demo sample: paragraphing, punctuation, a speaker label, fillers removed, and the unclear passage marked [inaudible] instead of guessed. Paste your transcript to get your version (sign-in required; one credit per run).

WHY YOU CAN TRUST IT

We put “no words added” into our product principles.

Not one added sentence

No explanations, transitions, or conclusions. If the speaker did not say it, it does not appear.

Figures and facts stay as spoken

Even a number that looks wrong is reproduced, then listed for you to confirm against the audio.

Unclear audio is marked, not guessed

Guessing turns muffled audio into a confident quotation. Marking it is the safe option.

Content is not stored

Submitted text is used for this single AI request and is not intentionally persisted. See the Privacy Policy.

FAQ

Ask these before you pay.

Will you delete anything that was said?

No. Only fillers and repeated openings go. No substantive content is removed, and nothing the speaker did not say is added.

What if a number looks wrong?

We keep it. Facts, figures, dates, and names are preserved verbatim; anything that looks off is listed for you to confirm against the audio.

How do you handle inaudible parts?

They are marked [inaudible] with their position so you can replay them. We will not guess a word that merely sounds right.

What does one cleanup cost?

One credit per run, usually done in seconds, returning the cleaned transcript, cleanup notes, and items to verify.

Do credits expire?

No. Credits never expire — whatever you do not use carries over to the next cycle instead of resetting.

Is there batch processing or an API?

Not yet. Each run processes one document and consumes one credit; there is no bulk upload or API today. Email us with your use case and we will assess it.

Is my content stored?

Not intentionally. Text is used for this single AI request only. See the Privacy Policy.

Which transcripts fit?

Meeting notes, user interviews, podcasts, lecture recordings, online talks — any spoken content already transcribed but not yet readable, in Chinese or English.

SIMPLE PRICING

Prove the value before you upgrade.

Starter

$9/ month

For an occasional publishing cadence, with a focused monthly credit allocation.

50 optimization credits per monthly renewal
Complete GEO analysis
Chinese and English output
Markdown export

Pro

Recommended$29/ month

For frequent publishers who iterate on content continuously.

250 optimization credits per monthly renewal
Complete GEO analysis
Chinese and English output
Markdown export

Advanced

$79/ month

Our largest monthly credit allocation, for the heaviest continuous usage.

1,000 optimization credits per monthly renewal
Complete GEO analysis
Chinese and English output
Markdown export

Credits never expire: unused credits carry over instead of resetting. Credits apply to this site only and are not shared with other sites.

Prices shown are USD baselines; final currency, taxes, and discounts are determined by Paddle for the checkout location.