README.md

metadata

license: apache-2.0
task_categories:
  - audio-classification
  - audio-to-audio
language:
  - en
tags:
  - audio
  - sound-separation
  - universal-sound-separation
  - audio-mixing
  - audioset
pretty_name: Hive Dataset
size_categories:
  - 10M<n<100M
dataset_info:
  features:
    - name: mix_id
      dtype: string
    - name: split
      dtype: string
    - name: sample_rate
      dtype: int32
    - name: target_duration
      dtype: float64
    - name: num_sources
      dtype: int32
    - name: sources
      sequence:
        - name: source_id
          dtype: string
        - name: path
          dtype: string
        - name: label
          dtype: string
        - name: crop_start_second
          dtype: float64
        - name: crop_end_second
          dtype: float64
        - name: chunk_start_second
          dtype: float64
        - name: chunk_end_second
          dtype: float64
        - name: rms_gain
          dtype: float64
        - name: snr_db
          dtype: float64
        - name: applied_weight
          dtype: float64
    - name: global_normalization_factor
      dtype: float64
    - name: final_max_amplitude
      dtype: float64
  splits:
    - name: train
      num_examples: 5000000
    - name: validation
      num_examples: 500000
    - name: test
      num_examples: 100000

A Semantically Consistent Dataset for Data-Efficient Query-Based Universal Sound Separation

Kai Li^*, Jintao Cheng^*, Chang Zeng, Zijun Yan, Helin Wang, Zixiong Su, Bo Zheng, Xiaolin Hu
Tsinghua University, Shanda AI, Johns Hopkins University
^*Equal contribution
Completed during Kai Li's internship at Shanda AI.
📜 Arxiv 2026 | 🎶 Demo

Usage

from datasets import load_dataset

# Load full dataset
dataset = load_dataset("ShandaAI/Hive")

# Load specific split
train_data = load_dataset("ShandaAI/Hive", split="train")

# Streaming mode (recommended for large datasets)
dataset = load_dataset("ShandaAI/Hive", streaming=True)

📄 Dataset Description

Hive is a high-quality synthetic dataset designed for Universal Sound Separation (USS). Unlike traditional methods relying on weakly-labeled in-the-wild data, Hive leverages an automated data collection pipeline to mine high-purity single-event segments from complex acoustic environments and synthesizes mixtures with semantically consistent constraints.

Key Features

• Purity over Scale: 2.4k hours achieving competitive performance with million-hour baselines (~0.2% data scale)
• Single-label Clean Supervision: Rigorous semantic-acoustic alignment eliminating co-occurrence noise
• Semantically Consistent Mixing: Logic-based co-occurrence matrix ensuring realistic acoustic scenes
• High Fidelity: 44.1kHz sample rate for high-quality audio

Dataset Scale

Metric	Value
Training Set Raw Audio	2,442 hours
Val & Test Set Raw Audio	292 hours
Mixed Samples	19.6M mixtures
Total Mixed Duration	~22.4k hours
Label Categories	283 classes
Sample Rate	44.1 kHz
Training Sample Duration	4 seconds
Test Sample Duration	10 seconds

Dataset Splits

Split	Samples	Description
Train	17.5M	Training mixtures (4s duration)
Validation	1.75M	Validation mixtures
Test	350k	Test mixtures (10s duration)

---

📂 Dataset Structure

Directory Organization

hive-datasets-parquet/
├── README.md
├── train/
│   └── data.parquet
├── validation/
│   └── data.parquet
└── test/
    └── data.parquet

Each split contains a single Parquet file with all mixture metadata. The num_sources field indicates the number of sources (2-5) for each mixture.

---

📋 Data Fields

JSON Schema

Each JSON object contains complete generation parameters for reproducing a mixture sample:

{
    "mix_id": "sample_00000003",
    "split": "train",
    "sample_rate": 44100,
    "target_duration": 4.0,
    "num_sources": 2,
    "sources": {
        "source_id": ["s1", "s2"],
        "path": ["relative/path/to/audio1", "relative/path/to/audio2"],
        "label": ["Ocean", "Rain"],
        "crop_start_second": [1.396, 2.5],
        "crop_end_second": [5.396, 6.5],
        "chunk_start_second": [35.0, 20.0],
        "chunk_end_second": [45.0, 30.0],
        "rms_gain": [3.546, 2.1],
        "snr_db": [0.0, -3.0],
        "applied_weight": [3.546, 1.487]
    },
    "global_normalization_factor": 0.786,
    "final_max_amplitude": 0.95
}

Field Descriptions

1. Basic Info Fields

Field	Type	Description
mix_id	string	Unique identifier for the mixture task
split	string	Dataset partition (train / validation / test)
sample_rate	int32	Audio sample rate in Hz (44100)
target_duration	float64	Target duration in seconds (4.0 for train, 10.0 for test)
num_sources	int32	Number of audio sources in this mixture (2-5)

2. Source Information (sources)

Metadata required to reproduce the mixing process for each audio source. Stored in columnar format (dict of lists) for efficient Parquet storage:

Field	Type	Description
source_id	list[string]	Source identifiers (s1, s2, ...)
path	list[string]	Relative paths to the source audio files
label	list[string]	AudioSet ontology labels for each source
chunk_start_second	list[float64]	Start times (seconds) for reading from original audio files
chunk_end_second	list[float64]	End times (seconds) for reading from original audio files
crop_start_second	list[float64]	Precise start positions (seconds) for reproducible random extraction
crop_end_second	list[float64]	Precise end positions (seconds) for reproducible random extraction
rms_gain	list[float64]	Energy normalization coefficients: $\text{target_rms} / \text{current_rms}$
snr_db	list[float64]	Signal-to-noise ratios in dB assigned to each source
applied_weight	list[float64]	Final scaling weights: $\text{rms_gain} \times 10^{(\text{snr_db} / 20)}$

3. Mixing Parameters

Global processing parameters after combining multiple audio sources:

Field	Type	Description
global_normalization_factor	float64	Anti-clipping scaling coefficient: $0.95 / \text{max_val}$
final_max_amplitude	float64	Maximum amplitude threshold (0.95) to prevent bit-depth overflow

Detailed Field Explanations

Cropping Logic

• chunk_start/end_second: Defines the reading interval from the original audio file
• crop_start/end_second: Records the precise random cropping position, ensuring exact reproducibility across runs

Energy Normalization (rms_gain)

Adjusts different audio sources to the same energy level: $$\text{rms\_gain}=\frac{\text{target\_rms}}{\text{current\_rms}}\backslash text\{rms\backslash \_gain\}\; =\; \backslash frac\{\backslash text\{target\backslash \_rms\}\}\{\backslash text\{current\backslash \_rms\}\}$$

Signal-to-Noise Ratio (snr_db)

The SNR value assigned to each source, sampled from a predefined range using random.uniform(snr_range[0], snr_range[1]).

Applied Weight

The comprehensive scaling weight combining energy normalization and SNR adjustment: $$\text{applied\_weight}=\text{rms\_gain}\times {10}^{(\text{snr\_db}\mathrm{/}20)}\backslash text\{applied\backslash \_weight\}\; =\; \backslash text\{rms\backslash \_gain\}\; \backslash times\; 10^\{(\backslash text\{snr\backslash \_db\}\; /\; 20)\}$$

This is the final coefficient applied to the original waveform.

Global Normalization Factor

Prevents audio clipping after mixing: $$\text{global\_normalization\_factor}=\frac{0.95}{\text{max\_val}}\backslash text\{global\backslash \_normalization\backslash \_factor\}\; =\; \backslash frac\{0.95\}\{\backslash text\{max\backslash \_val\}\}$$

Where max_val is the peak amplitude (absolute value) of the mixed signal.

---

🔧 Usage

Download Metadata

from datasets import load_dataset

# Load specific split and mixture type
dataset = load_dataset("ShandaAI/Hive", split="train")

Generate Mixed Audio

Please refer to the official GitHub repository for the complete audio generation pipeline.

# Clone the repository
git clone https://github.com/ShandaAI/Hive.git
cd Hive/hive_dataset

# Generate mixtures from metadata
python mix_from_metadata/mix_from_metadata.py \
    --metadata_dir /path/to/downloaded/metadata \
    --output_dir ./hive_dataset \
    --dataset_paths dataset_paths.json \
    --num_processes 16

---

📚 Source Datasets

Hive integrates 12 public datasets to construct a long-tailed acoustic space:

#	Dataset	Clips	Duration (h)	License
1	BBC Sound Effects	369,603	1,020.62	Remix License
2	AudioSet	326,890	896.61	CC BY
3	VGGSound	115,191	319.10	CC BY 4.0
4	MUSIC21	32,701	90.28	YouTube Standard
5	FreeSound	17,451	46.90	CC0/BY/BY-NC
6	ClothoV2	14,759	38.19	Non-Commercial Research
7	Voicebank-DEMAND	12,376	9.94	CC BY 4.0
8	AVE	3,054	6.91	CC BY-NC-SA
9	SoundBible	2,501	5.78	CC BY 4.0
10	DCASE	1,969	5.46	Academic Use
11	ESC50	1,433	1.99	CC BY-NC 3.0
12	FSD50K	636	0.80	Creative Commons
	Total	898,564	2,442.60

Important Note: This repository releases only metadata (JSON files containing mixing parameters and source references) for reproducibility. Users must independently download and prepare the source datasets according to their respective licenses.

---

📖 Citation

If you use this dataset, please cite:

---

⚖️ License

This dataset metadata is released under the Apache License 2.0.

Please note that the source audio files are subject to their original licenses. Users must comply with the respective licenses when using the source datasets.

---

🙏 Acknowledgments

We extend our gratitude to the researchers and organizations who curated the foundational datasets that made Hive possible:

• BBC Sound Effects - Professional-grade recordings with broadcast-level fidelity
• AudioSet (Google) - Large-scale audio benchmark
• VGGSound (University of Oxford) - Real-world acoustic diversity
• FreeSound (MTG-UPF) - Rich crowdsourced soundscapes
• And all other contributing datasets

---

📬 Contact

For questions or issues, please open an issue on the GitHub repository or contact the authors.