Upload a video or paste a supported media link. Whisper AI converts the spoken track into searchable text with timestamps, optional speaker labels, editing tools, and export formats for documents or captions.
Upload MP4, MOV, WebM, AVI, MKV, MP3, or WAV, or start with a supported public video link.
Direct answer
Video-to-text conversion separates the spoken audio from a supported upload or public media link and processes that audio with automatic speech recognition. Whisper AI returns an editable transcript with timed segments; optional diarization can group passages by detected speaker, while export controls create reading copies, documents, structured JSON, or SRT and VTT caption drafts. The current upload route accepts common video and audio formats up to the app's stated 1 GB limit. Transcription quality depends more on the recording than the picture: clear microphones, limited echo, low background music, and speakers who do not overlap generally produce a stronger first draft. Speaker labels identify voice clusters rather than real identities, so rename them during review. Before publishing, verify names, numbers, specialist vocabulary, and high-stakes statements against the recording. Upload only media you are authorized to process and apply appropriate access and retention rules to the finished transcript.
Upload original footage and private exports that you are authorized to process. Use a public link when the media is already accessible without credentials. Both routes lead to the same editor, but choosing deliberately prevents passwords, account sessions, and temporary private tokens from entering a link resolver.
Choose a safe video sourceClear speech matters more than picture resolution. If you control the file, reduce long silent sections, avoid clipped audio, and limit music or echo around important dialogue. Select the known language and supply specialist vocabulary when names, acronyms, or product terms matter.
Prepare the recordingEnable diarization for interviews, panels, podcasts, and calls. The model groups similar voices into speaker labels but does not know the participants' real names, so use playback and context to rename confirmed speakers during review.
Identify voice changesUse TXT for a reading copy, DOCX for collaboration, JSON for a software pipeline, and SRT or VTT when timing matters. Caption files still need a visual pass for line length, reading speed, and cut points inside the target editor.
Create the right outputScribe v2 supports more than 90 documented languages for transcription. Create the source-language transcript first, review material passages, and retain it beside any translated copy so editors can check names, terminology, and nuance.
Process multilingual videoUse a local media file or resolve a supported public video page.
Optional diarization separates dialogue in interviews, panels, and calls.
Keep transcript text aligned to useful points in the source recording.
Find passages and correct terminology before the transcript leaves the workspace.
Use Scribe v2 language selection or automatic detection across more than 90 documented languages.
Download document, plain-text, caption, and structured-data formats.
Choose a supported file from your device or enter a public media URL.
Select language, speaker labels, and the formatting style your content needs.
Search the result, correct details, and download text or timed subtitles.
The upload route gives you the most control over privacy, media quality, language settings, and speaker detection before the speech model begins.
Speech models perform best when voices are clear and close to the microphone. Excessive music, room echo, wind, clipped audio, and several people talking at once can reduce accuracy. If you control the edit, normalize speech and remove long silent sections before upload. You do not need a cinema-quality picture because the transcription provider primarily needs the spoken audio track.
Diarization adds processing cost and is useful for interviews, panels, podcasts, hearings, and calls. A single-narrator tutorial rarely needs it. Generated labels describe voice clusters rather than known identities, so review the result and rename speakers yourself. If one person changes microphones or the audio contains strong background voices, the model may create extra speaker labels that require correction.
Use TXT for a lightweight reading copy, DOCX for collaborative editing, JSON for software pipelines, and SRT or VTT when timing matters. Caption files should be reviewed inside the target video editor because line length, reading speed, and cut points affect accessibility. A technically valid subtitle file can still be uncomfortable to read if too much text appears at once.
Only upload media you are permitted to process. Recordings can contain customer information, internal strategy, personal data, or confidential conversations. Follow the policy that applies to the recording, restrict account access, and delete tasks that no longer have a business purpose. For regulated or high-stakes material, automatic speech recognition should be one controlled processing step rather than the final record.
Related resources
Upload private files, record in the browser, edit transcripts, and manage exports.
Transcription accuracy guidePrepare audio and review names, numbers, speakers, and specialist vocabulary.
Transcription pricingCheck current plan allowances, minute packs, and the free account allowance.
FAQ
Common formats including MP4, MOV, WebM, AVI, and MKV are accepted, along with many audio formats.
The current upload pipeline accepts media up to 1 GB. Available credits and recording duration also determine whether a task can start.
Yes. Turn on speaker labels before submission to use the diarization model, then rename labels in the finished transcript.
Yes. Signed-in users can keep an edited version, search it, restore recent revisions, and export the corrected result.
No. Public page URLs are resolved on the server, and the transcription provider reads the resulting media stream.
Upload a file or paste a link, then move from speech to searchable, editable, export-ready text in one workspace.
Start transcribing