iclr2027anon/synthetic-detect-diarize

Synthetic Detection and Diarization

This dataset contains 1,000,000 ten-second sound scenes with event and speaker annotations. WAV audio is included in data-20k/train-*.tar.

Quick start

from datasets import load_dataset

metadata = load_dataset("iclr2027anon/synthetic-detect-diarize", split="train")

Download and extract the TAR shards from data-20k/. Load a row's WAV from audio/<audio_name>.

Metadata

Column	Meaning
id	Scene ID: sed_diarization_ followed by 32 hexadecimal characters.
audio_name	WAV filename inside a TAR shard's audio/ directory.

IDs identify rows; filenames locate audio. ROOTS references these exact IDs in source_audio_ids.

Annotations

Each scene has a Raven-style selection table at selection_tables/<audio_name without .wav>_selection_table.txt in the same TAR. Times are in seconds within the scene. Source File identifies a Pseudovox clip: match its final filename to Pseudovox audio_name, then read its id. Speaker distinguishes sources within a scene. The table also records SNR and source references.

License

Source licenses and attribution requirements apply to the audio. See the repository's license metadata and original recording sources.