API Contract Coverage

This inventory defines the high-value public API contract for Phase 4. A listed concept must have a purpose, input/output constraints, side effects or exceptions, relevant serialization/configuration behavior, and a focused usage path in the linked documentation. It is intentionally not an inventory of all implementation modules.

The machine-readable source of truth is docs/api_inventory.json. Run python docs/check_api_coverage.py to verify that every classified concept still has an assigned existing page that mentions it; documentation CI enforces the same check.

Stable contracts

Namespace	High-value public concepts	Contract documentation
`hyperion.io`	Feature reader/writer factories; `DataReader`/`DataWriter`; Ark/HDF5 writers; sequential/random audio readers; VAD factories.	Foundation API Contracts, Input and Output API, and Prepare Dataset Metadata and VAD.
`hyperion.utils`	`InfoTable` and manifest subclasses; `HyperDataset`; dense and sparse trial tables; enrollment/class metadata; Kaldi compatibility helpers.	Foundation API Contracts, Hyperion Data Model, Working With Info Tables, Working With HyperDataset, and Working With Trial Tables.
`hyperion.metrics`	EER/DCF functions; verification, adversarial, anonymization, speech-quality, and VoxProfile evaluators.	Statistical, Evaluation, and Preparation API Contracts, Metrics and Evaluation API, and Extract, Score, and Evaluate X-Vectors.
`hyperion.data_prep`	`DataPrep` registration, parser integration, corpus-preparation output contract, and manifest validation.	Statistical, Evaluation, and Preparation API Contracts, Data Preparation API, and Prepare Dataset Metadata and VAD.
`hyperion.np`	`HyperNPModel` serialization; transforms; PLDA factory/scoring; calibration; adaptive S-Norm; array metrics.	Statistical, Evaluation, and Preparation API Contracts, NumPy Backend API, NumPy Backend Extension Points, and Save and Load Models and Backends.
`hyperion.torch`	`HyperTorchModel` serialization/train modes; waveform x-vectors; `NetArch` forward/mask contract; datasets/samplers; trainers; optimizers and schedulers; stable layers/blocks/narchs.	PyTorch API Contracts, PyTorch API Overview, PyTorch Extension Points, PyTorch Layers and Architecture Catalog, and PyTorch Training Support.
`hyperion.text_norm`	Basic/English text normalizers, English number normalizers, and spelling normalization behavior.	Statistical, Evaluation, and Preparation API Contracts and Text Normalization API.

Experimental or intentionally excluded surfaces

hyperion.torch.models.dac, VITS/freevc anonymization, transducers, and Q-vector models are documented as experimental in Experimental Components. Their APIs may be described for evaluation, but they are not a stable extension contract.

hyperion.helpers is not currently a supported public namespace. It is therefore deliberately excluded from this inventory rather than receiving an implicit API guarantee. If a helper becomes supported, add it to Public Surface Inventory, this table, and a contract page in the same change.

Review rule

When a change adds or materially changes a listed public concept, update its linked contract page in the same pull request. A generated signature alone is not sufficient: the documentation must state data layout, compatibility, and observable behavior needed by callers.