Proposal: URML (substrate-neutral robot intent) manifest declaration for whisper.cpp as the embedded listen substrate

[idoco2003]

Hi @ggml-org/whisper.cpp team,

Proposing a URML v0.1 capability-manifest mapping for whisper.cpp over ggml-org/whisper.cpp. URML (Apache-2.0) is a substrate-neutral spec for robot intent: typed primitive vocabulary + capability manifest + static validator.

whisper.cpp is the embedded substrate that closes URML's Layer-2 listen primitive on Layer-1 robots without Python — PX4 / NuttX, vendor RTOSes, microcontroller-class boards. URML's manifest needs a runtime-dependency-profile field (no_python / libc_only / rtos) that whisper.cpp's deployment profile is the canonical example of; the field generalizes to other embedded substrates URML will engage.

This is proposal-only, part of URML's Move #12 outreach (16 RFCs covering speech / translation / robot-command-library substrates for URML's NL layer).

Full RFC with manifest mapping, three alternatives, and the no-Python deployment design discussion: https://github.com/URML-MARS/URML/blob/main/docs/rfcs/0155-whisper-cpp-outreach.md

Questions worth maintainer input on:

1. STT-inference-runtime declaration. Would whisper.cpp benefit from URML's manifest declaring the runtime explicitly (stt_inference_runtime: ggml), or is this internal detail?
2. Runtime-dependency-profile declaration. Does the whisper.cpp team have a convention for documenting "this build requires libc only, no Python, no CUDA"? URML's stt_runtime_dependency_profile field would reflect that.
3. ggml-quantization declaration. Is the q4_0 / q4_1 / q5_0 / q8_0 / f16 / f32 enumeration the right granularity, or should it be a free-form ggml-quant-string?
4. Adapter home. URML-side adapter calling whisper.cpp via subprocess in reference/speech-bridge/, contributed example in whisper.cpp/examples/, or external bridge repo?
5. Org-migration handling. Is ggml-org/whisper.cpp the URL going forward, or should historical ggerganov/whisper.cpp references still be honored in URML's documentation?
6. Anything else.

Ido Yahalomi (URML maintainer, urml.dev, greenvh@gmail.com)

---

AI-assisted prose, maintainer-reviewed before posting (see VIBE.md). Human-only correspondence available on request.