BaseExtractor

Defined in: voxatlas.features.base_extractor

class voxatlas.features.base_extractor.BaseExtractor

Bases: ABC

Abstract base class for public VoxAtlas feature extractors.

Concrete extractors declare a feature name, the units they consume and produce, optional upstream dependencies, and a compute method. The pipeline uses this shared interface to execute every feature family through the registry.

Notes

Extractors should remain stateless. Dependency outputs should be read from feature_input.context["feature_store"] rather than stored on the extractor instance.

Examples

Usage example:

class MyExtractor(BaseExtractor):
    name = "custom.example"
    input_units = "token"
    output_units = "token"
    dependencies = []
    default_config = {}

    def compute(self, feature_input, params):
        ...

name: str

input_units: str | None = None

output_units: str | None = None

dependencies: list[str] = []

default_config: dict = {}

abstractmethod compute(feature_input, params)

Compute the extractor output for one stream.

Parameters:

• feature_input (FeatureInput) – Prepared input bundle containing audio, unit tables, and pipeline context.

• params (dict) – Resolved configuration for the extractor.

Returns:

Structured VoxAtlas feature output.

Return type:

object

Notes

Implementations should raise informative errors when required modalities or dependency outputs are unavailable.

Examples

Usage example:

output = extractor.compute(feature_input, params)
print(output)