Expand description
Durable, automatic audio transcription and speaker correction.
AudioIngress accepts owned WAV bytes, persists them before returning,
and automatically transcribes, classifies, and reconciles them in the
background.
Structs§
- Audio
Chunk Request - One unstructured audio request delegated to the application intelligence router.
- Audio
Ingress - Cloneable handle to durable, automatically processed audio.
- Audio
Input - Owned audio and its source metadata.
- Audio
Transcriber - Complete audio workflow backed by caller-owned typed intelligence operations.
- Candidate
Mapping - Classifier evidence retained for one chunk-local speaker.
- Chunk
Confirmation - Exact speaker resolutions and signoff for one review chunk.
- Correction
Chunk - One complete chunk in a recording-level correction packet.
- Correction
Observation - One deterministic classifier observation in a correction packet.
- Correction
Packet - Complete transport-neutral correction packet for one recording.
- Error
- Error returned by the AudioIngress API.
- Intelligence
Error - Sanitized intelligence-router failure returned to the audio workflow.
- Observation
Confirmation - One observation-level human resolution.
- Observation
Key - One caller-owned observation identity.
- Parsed
Chunk - Complete normalized speaker structure for one raw Gemini result.
- Parsed
Speaker - One typed speaker row normalized from a raw Gemini response.
- Recording
Status - Current state, immutable identity, and optional completed correction packet.
- Speaker
Review Audio - One exact source interval encoded for human speaker review.
- Status
- Complete current library state.
- Step
Error - Sanitized terminal or retryable step error.
- Step
Status - Current status of one ordered pipeline step.
- Submission
- Result of durably submitting audio.
- Text
Generation Request - One tool-free text request delegated to the application intelligence router.
- Transcription
Job - Cloneable handle for polling one in-memory transcription.
- Transcription
Status - Cheap cloneable snapshot returned by
TranscriptionJob::status.
Enums§
- Confirmation
State - Durable identity-confirmation lifecycle for a correction packet.
- Error
Kind - Stable library error category.
- JobState
- Overall state of an in-memory transcription job.
- Legacy
Review Disposition - Durable resolution for a finalized recording with an obsolete, unsigned packet.
- Recording
State - Automatic processing state of one recording.
- Speaker
Resolution - Human resolution for one chunk-local speaker.
- Step
- One ordered pipeline operation.
- Step
State - State of one pipeline step.
Constants§
- CLASSIFIER_
MODEL - Exact classifier model cohort component.
- CLASSIFIER_
PROMPT_ VERSION - Exact classifier prompt-version cohort component.
- CLASSIFIER_
PROVIDER - Exact classifier provider cohort component.
- CLASSIFIER_
SCHEMA_ VERSION - Exact classifier feature-schema cohort component.
- RECONCILIATION_
MODEL - GPT model used to parse and reconcile chunk results.
- RECONCILIATION_
REASONING - GPT reasoning setting used for parsing and reconciliation.
- TRANSCRIPTION_
MODEL - Gemini model used for chunk speaker analysis.
Type Aliases§
- Audio
Chunk Call - Typed unstructured-audio call implemented by the application intelligence router.
- Feature
Row - The frozen 24-value feature row used by the classifier.
- Intelligence
Future - Sendable future returned by an injected intelligence call.
- Text
Generation Call - Typed tool-free text call implemented by the application intelligence router.