speechlab

Linux-native lab workspace for speech, audio, and ASR-adjacent feature experiments for SocioProphet services.

Role in the Holmes lab family

speechlab is the first governed pilot lab for the Holmes speech-intelligence stack. It emits a functional-service.v1 manifest that can be consumed by:

• SocioProphet/holmes as the speech-intelligence product surface;
• SocioProphet/functional-model-surfaces as the normative service contract spine;
• SocioProphet/sociosphere as workspace registry and maturity controller;
• SourceOS-Linux/sourceos-model-carry as disabled-by-default SourceOS carry metadata;
• SociOS-Linux/socios as opt-in training/tuning orchestration;
• SocioProphet/model-governance-ledger and SocioProphet/model-router for promotion and routing.

This repository does not store model weights, mutable adapters, personal data, secrets, or promotion authority.

Local validation

make validate
make smoke
make carry

The smoke path is deterministic and offline-only.