speechlab

Status: recovery scaffold
Authority role: audio-first review and spoken-corpus experimentation surface
Claim level: product/research infrastructure planning; no runtime model commitment

speechlab is the SocioProphet lab surface for audio-first human-agent review, spoken corpus handling, listening-mode workflows, transcript provenance, and speech-channel confusability testing.

This repository is not currently the canonical speech model runtime, transcription service, or production audio ingestion plane. It is the place to define how audio-first review should be structured before downstream systems adopt it.

Start here

• docs/audio-first-review-runtime-v0.md — recovery contract for audio-first review/runtime semantics.

Intended consumers

• prophet-platform — workroom listening/review UX and audio-friendly sectioning.
• alexandrian-academy — spoken teaching objects and listenable lessons.
• systems-learning-loops — review patterns, spoken postmortems, and training loops.
• ontogenesis — audio anchors, transcript selectors, provenance, and semantic promotion boundaries.
• model-governance-ledger — transcription, summarization, evaluation, drift, and audio-processing receipts.
• agentplane — agent review actions, operator readouts, and receipt-backed effectful workflows.

Non-goals

This scaffold does not select ASR/TTS models, define production storage, ingest private audio, train speech models, or authorize durable memory formation from spoken material.

Any future runtime integration must respect DoNotLearn / DoNotLink, transcript provenance, human review boundaries, and receipt discipline.