datasets
Training and evaluation data, with the modality, task and licence stated up front. Listed live from the Hugging Face Hub.
knnnadi2026-adi20-micro-25pct-knnvc
NADI 2026 ADI20-micro — kNN-VC augmented (4 target voices)
Voice-converted copy of the 25% stratified subset (seed 42) of
UBC-NLP/NADI_2026_ADI20_micro, made with kNN-VC
following Abdullah et al. 2025.
Configs: voice_01–voice_04, 16,757 rows each, train split only.
Validation/test audio is deliberately left natural.
Target voices: 4 Arabic speakers from Common Voice (~60s each), gender-balanced,
the same set used across all dialects.
column
meaning
audio
converted… See the full description on the dataset page: https://huggingface.co/datasets/nadi-task2/nadi2026-adi20-micro-25pct-knnvc.Mega-Fast-KNN-Captioningprotea-lafa-knn-v227
PROTEA-KNN — v227 reference bundle
Frozen, inference-only reference data for the protea-knn-v1 LAFA
submission. Lets anyone reproduce PROTEA's simple KNN baseline without any
PROTEA infrastructure (no database, no queues): mount this bundle into the
container and run.
Embedding model: Rostlab/prot_t5_xl_half_uniref50-enc (mean pooled,
1024-dim) — the same model as LAFA's own ProtT5 baseline. The difference is
the data: PROTEA's reference protein set.
Cutoff: GOA release v227… See the full description on the dataset page: https://huggingface.co/datasets/XaxiPiruli/protea-lafa-knn-v227.knn-prompt-datastoreeval-pack-uwc-of-my-project-knn0x1-be652975
Eval pack uwc of my Project knn0x1
detect carpet
This dataset mirrors public data-pack render outputs from Physicl.
Each row represents one render view. The image column contains a stable URL to the primary render image uploaded under /data; image_path stores the relative repository path and data_commit_sha pins the Hugging Face dataset commit used by those URLs. Files are uploaded as downloaded unless optional PNG recompression is enabled by the sync operator.
Additional render… See the full description on the dataset page: https://huggingface.co/datasets/physicl-community/eval-pack-uwc-of-my-project-knn0x1-be652975.orpheus_tts_knn_vc_pathological
Combined Orpheus TTS 3B + Same-Speaker KNN Voice Conversion Dataset
Dataset Overview
This dataset contains fine-tuned Orpheus TTS 3B synthetic speech enhanced with same-speaker KNN voice conversion a
Enhancement Method: Orpheus TTS 3B LoRA fine-tuned synthetic speech → Same-speaker KNN voice conversion using real audio references from the same speaker
Dataset Statistics
Total Samples: 759
Total Duration: 2583.38 seconds (43.06 minutes)
Speakers: 8 speakers… See the full description on the dataset page: https://huggingface.co/datasets/resproj007/orpheus_tts_knn_vc_pathological.spark_tts_knn_vc_pathological
Dataset Overview
Total Samples: 785
Total Duration: 3006.46 seconds (50.11 minutes)
Speakers: 8 speakers
Corpora: TORGO, UA-Speech, LibriSpeech
Sample Rate: 16kHz (KNN-VC output rate, native Spark compatibility)
Audio Format: WAV
all-nli-knn-hard-negativesknn-prompt-datastoreknndatasesame_tts_knn_vc_pathological
Dataset Overview
Total Samples: 759
Total Duration: 2892.82 seconds (48.21 minutes)
Speakers: 8 speakers
Corpora: TORGO, UA-Speech, LibriSpeech
Sample Rate: 16kHz (KNN-VC output rate)
Audio Format: WAV
TR-Law-tre5-knn-45KmHuBERT-knnvc-datasetdata-synthetic-synthetic-vault-to-bcb-by-az-2k-descr-knnpathological_knn_vc
Combined Same-Speaker KNN Voice Conversion Dataset
Dataset Statistics
Total Samples: 799
Total Duration: 0.4 hours
Speakers: 8
Corpora: TORGO, UA-Speech, LibriSpeech
Audio Format: 16kHz WAV
Baseline TTS: resproj007/baseline_orpheus_3b
Speaker Breakdown
Speaker
Name
Corpus
Condition
Gender
Samples
Duration
FC02
TORGO Healthy Female
TORGO
Healthy
Female
100
146.0s
M04
UA-Speech Male
UA-Speech
Dysarthric
Male
200
245.7s
M02
TORGO Dysarthric Male… See the full description on the dataset page: https://huggingface.co/datasets/resproj007/pathological_knn_vc.