ttsdb-core 0.1.4

ttsdb-core

ttsdb-core provides the shared “adapter” surface that every model package implements.

Key pieces

• VoiceCloningTTSBase: abstract base class for voice-cloning style TTS adapters.
  • Subclasses implement _load_model() and _synthesize() and use synthesize() for the public API.
• ModelConfig: helper for loading a model package’s bundled config.yaml (supports editable installs too).
• Vendoring utilities: helpers to make upstream research code importable at runtime:
  • setup_vendor_path(package_name): prepends src/<pkg>/_vendor/source[/code.root] to sys.path
  • vendor_context(...): optional context manager for research code that assumes a working directory or env vars

Model package conventions

Each model package lives in models/<name>/ and typically includes:

• pyproject.toml: packaging metadata (including requires-python)
• config.yaml: model metadata + dependency constraints + pointers to upstream code/weights
• src/ttsdb_<name>/__init__.py: the adapter implementation
• tests/: lightweight unit tests and optional integration tests (weights required)

Python version constraints

Because upstream research code often depends on specific Python versions, model packages express supported Python versions in two places:

• pyproject.toml → requires-python (PEP 440 specifier string)
• config.yaml → dependencies.python (same specifier string)

For local development automation, config.yaml may also include:

• dependencies.python_venv: a concrete interpreter version used by just setup when creating a venv
• dependencies.python_tested: optional explicit list of tested interpreters (human-facing)