datasets
Training and evaluation data, with the modality, task and licence stated up front. Listed live from the Hugging Face Hub.
artifactbench
ArtifactBench — AI-Generated Music Detection Benchmark
ArtifactBench v2 — lineage-aware frozen protocol
ArtifactBench v2 adds a metadata-first, lineage-aware evaluation protocol while
preserving the v1 and v1.1 releases below. Its frozen primary cohort contains
828 entries (605 AI-generated and 223 real) across 15 source strata, split by
content lineage into calibration, validation, and sealed-test partitions before
final model comparison.
The v2 package is under… See the full description on the dataset page: https://huggingface.co/datasets/intrect/artifactbench.swamiji-artifact-abc-grid
Where does the end-of-clip artifact come from?
Reported symptom: short polite replies "mess up at the end", and a "huge high"
is audible after the words finish — sometimes even when the sentence ends in a
full stop. Three candidate causes: the model, the streaming, or the Opus codec.
Each phrase here is generated twice — ending in ! and ending in . — and
each generation is rendered three ways, so exactly one variable moves at a
time. Six players per row.
rendering
what it… See the full description on the dataset page: https://huggingface.co/datasets/sw-voice/swamiji-artifact-abc-grid.swamiji-tail-artifact-clips
Reported end-of-clip artifact: the raw clips
The exact phrases reported as having "a weird sound at the end", straight out of
the production engine, three independent draws each. No processing at all —
this is what the model produced.
phrase
text
doingwell
I am doing well, thank you.
hereif
also I am here if you need anything.
hereforyou
I am here for you.
welcome
You are very welcome!
glad
Glad you think so!
scope
I can only answer about Sri Ganapati… See the full description on the dataset page: https://huggingface.co/datasets/sw-voice/swamiji-tail-artifact-clips.ArtifactBalladVoiceover