datasets
Training and evaluation data, with the modality, task and licence stated up front. Listed live from the Hugging Face Hub.
CaptionStew10M-Qwen3Omniall-ceos-audio-caption-combined
oridror/all-ceos-audio-caption-combined
MYD multi-speaker Hebrew TTS training set for Orpheus-3B.
Sourced from 6-ai/training/datasets/multimodal/audio_caption/ —
Hebrew transcripts paired with synthetic edge-tts audio (Hila female,
Avri male). License: Apache-2.0.
Schema
Each row in train.jsonl:
{
"audio_path": "<ceo>__<NNNNN>_<hash>.mp3",
"transcript": "<utf-8 hebrew>",
"speaker_id": "hila | avri",
"ceo": "adel | rafael | antonio | lavi | noa"… See the full description on the dataset page: https://huggingface.co/datasets/oridror/all-ceos-audio-caption-combined.fma-dataset-aug-caption
FMA-CLAP Caption Augmentation Dataset
Overview
This dataset is an enhanced version of the FMA (Free Music Archive) dataset, where we have augmented the original metadata with natural language captions generated using the CLAP (Contrastive Language-Audio Pretraining) model. The captions describe the genre, style, mood, and instrumentation of each track, making it more suitable for zero-shot learning, music classification, and text-to-music generation tasks.… See the full description on the dataset page: https://huggingface.co/datasets/solbon1212/fma-dataset-aug-caption.