metadata.json.txt

{
  "name": "OpenSER-700K",
  "description": "OpenSER-700K is a large-scale, high-quality speech emotion recognition dataset containing 700,258 audio samples across 7 emotion categories (Angry, Disgust, Fear, Happy, Neutral, Sad, Surprise). It is curated for deep learning, multimodal fusion, and benchmarking in affective computing and human-computer interaction research.",
  "version": "1.0.0",
  "total_audio_files": 700258,
  "duration": "≈1200 hours (avg. 6 seconds per clip)",
  "sampling_rate": 16000,
  "audio_format": "wav",
  "classes": {
    "angry": 98477,
    "disgust": 98477,
    "fear": 98477,
    "happy": 98477,
    "neutral": 98505,
    "sad": 98477,
    "surprise": 99348
  },
  "modalities": ["audio", "speech"],
  "recommended_usage": [
    "Training speech emotion recognition (SER) models",
    "Feature extraction with HuBERT, Wav2Vec2, or OpenL3",
    "Multimodal fusion with facial emotion embeddings",
    "Benchmarking speech-based affect recognition tasks"
  ],
  "dataset_structure": [
    "OpenSER-700K/{emotion}/{filename}.wav",
    "Metadata file: data.json (maps file path → emotion label)"
  ],
  "license": "cc-by-nc-4.0",
  "usage_notes": "This dataset is for academic and non-commercial research only. Please provide attribution when using it in publications or derivative works.",
  "source_datasets": [
    {"name": "RAVDESS", "license": "CC BY-NC"},
    {"name": "SAVEE", "license": "Research-use-only"},
    {"name": "TESS", "license": "Research-use-only"},
    {"name": "CREMA-D", "license": "CC BY-NC"},
    {"name": "EMO-DB", "license": "Research-use-only"},
    {"name": "IEMOCAP", "license": "Research-use-only"}
  ],
  "libraries": ["datasets", "torchaudio", "transformers"],
  "formats": ["wav"],
  "tags": ["speech-emotion-recognition", "audio", "SER", "emotion", "deep-learning"],
  "citation": "Adhavan M. OpenSER-700K: Large-Scale Speech Emotion Recognition Dataset. 2025. URL: https://huggingface.co/datasets/imadhavan/OpenSER-700K"
}