Journalistic file work is a chain of custody problem: get the interview into text, cut the audio you will publish, read the documents, strip what identifies a source, then lock what must stay confidential.
The order is transcribe, cut, read, strip, encrypt. Strip metadata before encrypting, because an encrypted file has to be unlocked before any other tool can touch it. And know which step leaves your machine: transcription does, everything else on this page does not.
1. Transcribe the interview
Video to Subtitles turns an audio or video file into a transcript with timecodes, and it can label speakers for a two-person interview. You can correct names, numbers and wording in the preview before downloading SRT, VTT or plain text. A free run covers 2 minutes per day, a free account receives 5 credits a day, and the upload limit is 30 MB per file, one file at a time.
This is the only step on the page that leaves your machine. The file is uploaded to a server and forwarded to Volcengine AI MediaKit for recognition. For a confidential source that is a real problem, and the honest answer is to transcribe only material that is safe to upload, or to use a local speech-to-text tool instead. The recogniser also covers Simplified Chinese and English only.
Video to Subtitles
Transcript with timecodes and speaker labels. Uploads the file and uses Cloud AI credits.
2. Cut the quote you will publish
Audio Trimmer draws the waveform and lets you drag the region edges or type exact times. It re-encodes the selection rather than copying compressed frames, so the cut lands where you asked rather than on the nearest frame boundary, and a fade of up to 5 seconds softens both ends. It opens MP3, WAV, M4A, AAC, OGG, FLAC and Opus, and exports MP3 or WAV.
Choose WAV if the clip still has to be edited or processed further; MP3 at 192 kbps is fine for publishing. One limit: it handles one region per file, so five quotes mean five passes. Nothing is uploaded, which makes it the safe place to work with raw interview audio.
Audio Trimmer
Cut a quote on the waveform with a fade of up to 5 seconds, all on your device.
3. Read the documents
PDF to Text handles both kinds of paperwork a story depends on. A court filing or a report exported from a word processor carries a real text layer, and that text is extracted instantly and exactly. A scanned letter is rendered and recognised with OCR instead. The result panel says how many pages came from each route, which tells you where to proofread.
English is built in, and eleven more languages can be selected, including Chinese, Japanese, Korean, German, French and Spanish. Everything runs locally, so a leaked document does not become more leaked by being read. Two limits: handwriting is not reliable, and the output is text, not structure, so a table comes out as lines rather than cells. An encrypted PDF has to be unlocked first.
PDF to Text
Read the text layer directly and run OCR on scans, all in the browser.
4. Strip what identifies a source
Remove EXIF cuts metadata out of JPEG and PNG without re-encoding the pixels. It shows GPS coordinates before you commit, and it re-reads the output afterwards to confirm that no metadata segments remain. The default mode keeps the ICC colour profile, so colours still look right when the photo is published.
PDF Metadata clears the Title, Author, Subject, Keywords, Creator and Producer fields that readers show under document properties. That is the field that often carries a journalist's own name or a newsroom laptop name into a published file.
Neither tool is redaction. A name or an address printed on the page is still there, and the PDF tool does not reliably remove an XMP packet written by a scanner. If a document identifies a source, edit the source and export a clean copy instead of trusting a strip pass. HEIC and WebP are not supported by the metadata tool.
Remove EXIF
Strip GPS and device data from JPEG and PNG without touching the pixels.
5. Lock what must stay confidential
Protect PDF writes AES-256 encryption in the browser. Set an open password, an optional owner password for permissions later, and decide whether the recipient may print, copy text and edit. The password and the file never leave your machine, so there is no server-side copy to leak.
Two limits are worth planning around. There is no password recovery, so a lost password loses the file for everyone, and the password should travel through a different channel than the file. And the permission switches are enforced by the reader rather than by cryptography, so treat them as a clear signal rather than a lock. Audio and video cannot be encrypted by these tools at all; for a sensitive recording, use a dedicated encrypted archive.
Protect PDF
AES-256 encryption with print, copy and edit permissions, entirely local.
What this chain cannot do
- Transcription uploads the file. Video to Subtitles sends the recording to a server, caps it at 30 MB, and recognises Chinese and English only.
- No automatic redaction. Metadata stripping does not hide a name or a face printed in the document, and no tool here draws a black box over text.
- No integrity check. Nothing produces a hash, a signature or a timestamp, so these tools cannot prove a file was not altered.
- PDF is the only format that can be encrypted. There is no password protection for audio, video or images.
- No chain of custody. Reading, cutting and encrypting leave no audit trail.
- HEIC and WebP cannot be stripped. Remove EXIF covers JPEG and PNG only.
- Nothing recovers lost detail. Trimming, compressing or re-encoding is one-way for the material that is cut or discarded.