Speech to Text

Clear speech with limited background sound gives the model a more useful source. Overlapping speakers, loud music and indistinct words can reduce accuracy. Listen to the source and proofread names, numbers and technical terms in the transcript.

Speech to text returns timed segments and retains word alignment when the provider supplies usable timing. Some segments may only have a start and end time. Correcting a segment's words removes that segment's old word alignment while retaining its segment timing.

Yes. Choose Identify speakers automatically, or a count from 2 to 6, and each line is labeled with who is speaking. Turn on Include sound notes to mark sounds such as [laughter] or [applause] where they happen; with sound notes on, speakers are found automatically. Review recordings where voices overlap carefully before using the transcript.

Use TXT for notes or a document, SRT for subtitle editors, and VTT for web captions. JSON keeps the language, text, timed segments and any remaining word alignment for software that can read structured data.

Corrections remain in the current editor and are included in copied text and downloads. They do not overwrite the original transcription in history. Export your revised copy before leaving the page.

The music-video editor uses the same transcription and review controls. Apply a transcript to that video's audio to replace its caption timeline, then review the timing before rendering. A transcript from different audio is not applied automatically.

Speech to Text draws 4 credits per run.