torch-dae: an AI skill-based framework for Audio Embedding Models

Status: pre-release · Version: 0.1.0

torch-dae is a model-agnostic control plane and AI skill-based workflow for evidence-grounded onboarding and integration of audio embedding models. It records source provenance, checkpoint identity, isolated environment resolution, wrapper behavior, embeddings, and runtime observations as explicit, validated artifacts.

No model-specific integrations are distributed in the current release. Model support is added through the canonical onboarding workflow and isolated model-specific environments.

Installation

Install the package-index distribution:

pip install torch-deepaudioembedding
torch-dae --help

Or install the source checkout:

git clone https://github.com/StefanoGiacomelli/torch_dae.git
cd torch_dae
uv sync --all-groups
uv run torch-dae --help

Start with the Quickstart, then follow the Audio-model onboarding for the skill workflow. Package users can consult the hand-curated Curated generic API; contributors should begin with Architecture and Contributing.

Getting started

• Installation
  • Requirements
  • Package index
  • Source checkout
• Quickstart

Tutorials

• Audio-model onboarding
• Agent interaction

User guide

• Model registry
• Environments
• Checkpoints
• Embeddings
• Model cards

Skill reference

• Skill overview
  • Analyze mode
  • Resolve-environment mode
  • Integrate mode
  • Verify mode
  • Card mode

API reference

• Curated generic API
  • Registry
  • Model execution
  • Outputs and embeddings
  • Capabilities and public errors
  • Checkpoint management
  • Model-card contracts
  • Environment specifications
  • Environment lifecycle
  • Runtime verification
  • Onboarding contracts
  • Static onboarding inspection
  • Report rendering
  • Recurring terms

Development

• Architecture
• Contributing
• Testing
• Documentation
• Releasing
  • Release sequence
  • GitHub environment: pypi
  • GitHub environment: testpypi
  • Read the Docs
  • Codecov
  • Zenodo
  • Branch protection

Reference

• CLI reference
• Schema reference
• Lifecycle reference

Citation

Cite the software entry in the repository’s CITATION.cff when citing the repository or package. Also cite the IEEE ISCC paper (DOI 10.1109/ISCC65549.2025.11326439) when discussing the framework design, standardization rationale, or deployment methodology.

Funding

Research project: Methods of Computational Auditory Scene Analysis and Synthesis supporting eXtended and Immersive Reality Services.

Research activities were mainly funded under the Ministerial Decree (DM) 118/2023, Mission 4, Component 1, Investment 4.1 of the National Recovery and Resilience Plan (PNRR) – “PNRR Research” – CUP: E11I23000100001.

Contact

Stefano Giacomelli
ICT - Ph.D. Candidate
Department of Information Engineering, Computer Science and Mathematics (DISIM)
University of L’Aquila, Italy

Email · GitHub · ORCID · Google Scholar · LinkedIn