AI · Document & photo interpretation

Speech transcription

Transcribe supported audio into timestamped source-linked segments.

Workflow

1. Request transcriptionSelect an authorized audio source and expected transcript schema.
2. Validate audio and modelRefuse unsupported or unauthorized sources before model work.
3. Transcribe segmentsProposed model work returns source-time-linked segments and abstention.
4. Return unavailableThe workflow returns an explicit unavailable or refused outcome without inventing content or mutating unrelated records.
5. Return transcriptThe proposed result preserves timestamps, confidence and source identity.

Failure boundaries

• No workerUnavailable: Do not fabricate text
• Unsupported formatRefused: Convert explicitly
• No timestampsInvalid result: Reject

ai.read.speech