EMMA-SC-SEP v1¶
Warning
Pre-implementation. This page describes proposed contracts. Class signatures, parameter types, schema fields, and behavior are subject to change before code lands. Once implementation exists, content here will be regenerated from docstrings or sourced from running tests.
Dataset. EMMA-SC-SEP-v1 covers the task, target domain, input, labels, splits, and the download command; this recipe pins only the rfgen configuration. Source. rfgen (primary).
rfgen recipe¶
rfgen commit:
<rfgen-commit>(frozen at release; placeholder until pinned).Emitters: the rfgen
emitter_zoogroup, a heterogeneous mix ofcomms,drone_rf, and other families, placed densely so scenes contain overlapping emitters. Emitter counts and overlap come from the rfgenscenedensity andplacementgroups.Receiver array: 8-element ULA (uniform linear array), half-wavelength spacing (
spacing_lambda: 0.5), via the rfgenMultiRXConfig.geometrymemberULA. Loaded into ArraySpec at load time.Channel environment(s):
E_URBANandE_RURAL, pinned through the rfgenchannelgroup.Labeler + label extraction: the rfgen
labelgroup with segmentation enabled (name: segmentationorjoint) writes per-emitter masks; LabelExtractor maps the segmentation output to per-emitter separation masks. The metric is SI-SDR (scale-invariant signal-to-distortion ratio).SNR (signal-to-noise ratio) regime: drawn from the rfgen
channel.snr_db_range.Sample counts:
<to be pinned at v1>.Splits + seeds:
train / dev / holdout seed ranges pinned at release; the holdout is a frozen secret seed, never published. OOD axis: EMITTER_MIX.
Real-capture counterpart¶
EMMA-REAL-OOD-v0.1 (fidelity reference where overlapping captures exist).
Notes¶
Separation needs the rfgen component-signal preservation (per-emitter signals kept distinct through the channel chain) so the mask labels are well-defined. The recipe depends on that preservation being correct.
A dedicated separation readout head is proposed to replace the interim classification head at v1.
References¶
Le Roux, Wisdom, Erdogan, and Hershey, “SDR: Half-baked or Well Done?,” ICASSP 2019, arXiv:1811.02508. SI-SDR for source separation. (verify)
See Also¶
Task reference: the
E-SC-SEPcontract.Datasets: the dataset registry.
Dataset recipe: the recipe schema this page pins.
Datasets validation: the recipe realism audit.