badc.chunking

Chunk planning helpers plus probe/aggregation logic.

See notes/chunking.md for the longer-term Plan A (HawkEars-driven probes, chunk overlap heuristics, manifest hashing, etc.). The functions below stay lightweight so CLI/tests can exercise the workflow while the more advanced algorithms are being built.

Functions

aggregate_detections(detections)	Aggregate placeholder detections by chunk prefix.
iter_chunk_placeholders(audio_path, ...)	Yield placeholder chunk identifiers for documentation/testing.
plan_chunk_ranges(duration_s, chunk_duration_s)	Return evenly spaced ranges that cover duration_s seconds.
probe_chunk_duration(audio_path[, ...])	Estimate a feasible chunk duration for audio_path based on GPU VRAM heuristics.
run_inference_on_chunks(chunk_ids)	Return mocked detection IDs for each chunk_id.
write_manifest(audio_path, chunk_duration_s, ...)	Write a chunk manifest CSV (placeholder hashing).

Classes

ChunkProbeAttempt(duration_s, ...)	Single attempt recorded while searching for a safe chunk duration.
ChunkProbeResult(file, max_duration_s[, ...])	Represents the outcome of a chunk-size probe run.