Redaction now extends to audio, completing the multimodal set: documents, images, and recordings all run through the same detection pipeline.
A recording is transcribed, the transcript is scanned by the same recognizers that handle documents, and the spans that match are muted in the audio and removed from the transcript. Call recordings, interviews, and voicemail can be retained without keeping the details spoken aloud in them.
Features
- Redaction across WAV, MP3, M4A, FLAC, and OGG
- Speech-to-text with word-level timestamps
- Detection over the transcript using the document recognizers
- Matched spans muted in the audio and removed from the transcript
- Both the redacted audio and the redacted transcript returned
Improvements
- Word-level timing keeps mutes tight, so surrounding speech survives
- Speaker separation, so a detection is attributed to who said it
- Long recordings are processed in chunks with progress reported per job
Getting Started
Audio uses the same endpoints as documents and images. See the API documentation for supported formats and transcription options.