| --- |
| license: other |
| task_categories: |
| - audio-classification |
| language: |
| - en |
| - hi |
| tags: |
| - deepfakes |
| pretty_name: Reality-aware SF |
| --- |
| # Reality-Aware Audio SceneFake Dataset |
| --- |
|
|
| ## Authors |
|
|
| - Jiya Sinha, Aarthi S, Akshay Agarwal, and Nalini Ratha |
| <!-- - Email id: {jiya22, saarthi24, akagarwal}@iiserb.ac.in --> |
| - Trustworthy BiometraVision Lab, IISER Bhopal, India |
| - University at Buffalo, USA |
|
|
|
|
| ## Overview |
| This dataset is designed for the study of **SceneFake** audios. It contains real-world recordings collected across diverse acoustic environments, with the goal of evaluating the limitations of synthetic scene-mixing approaches and enabling robust cross-dataset generalization. |
| It contains **15.42 hours of audio** recorded across **9 real-world acoustic scenes** in **Hindi and English**. |
| Unlike prior work that relies on synthetic mixing of clean speech and background noise, this dataset captures **naturally occurring interactions between speech and environment**. |
|
|
|
|
| ## Access |
| To get access to the dataset, submit the completed and duly signed [Agreement](https://github.com/sinhajiya/Real-SceneFake.git) via email to akagarwal@iiserb.ac.in. |
|
|
|
|
| ## Baselines |
|
|
| Baseline models and evaluation pipelines are available here: |
|
|
| **GitHub Repository:** |
| [https://github.com/sinhajiya/Real-SceneFake.git](https://github.com/sinhajiya/Real-SceneFake.git) |
|
|
|
|
| --- |
|
|
|
|
| ## Dataset Summary |
|
|
| | Feature | Value | |
| | ----------------------- | ----------------------- | |
| | Total Duration | 15.42 hours | |
| | Total Utterances | 2,190 | |
| | Languages | Hindi, English | |
| | Speakers | 47 (30 male, 17 female) | |
| | Age Range | 17–25 years | |
| | Acoustic Scenes | 9 | |
| | Recording Device | iPhone 15 | |
| | Utterance Length | 10–30 seconds | |
| | Sentences per Utterance | 5–6 | |
|
|
| --- |
|
|
| ## Data Collection Pipeline |
|
|
|
|
|  |
|
|
| * Recorded in **real-world, unconstrained environments** |
| * No studio setup or controlled SNR conditions |
| * Natural variation in: |
|
|
| * background noise |
| * speaker loudness and prosody |
| * environmental dynamics |
|
|
| Participants come from multiple Indian states with diverse **mother tongues**, introducing realistic **accent and pronunciation variability**. |
|
|
| All participants provided **informed consent** for data collection and usage. |
|
|
| --- |
|
|
| ## Dataset Structure |
|
|
| ``` |
| root/ |
| ├── real/ |
| ├── fake/ |
| ├── protocols/ |
| ├── pretrained_models/ |
| ├── SceneFakeAudio_info.csv |
| ├── SceneInfo.csv |
| ├── SpeakerInfo.csv |
| └── README.md |
| ``` |
|
|
| --- |
|
|
| ## Directory Details |
|
|
| ### `real/` |
|
|
| Contains **authentic recordings** collected directly from real-world acoustic environments. |
|
|
| --- |
|
|
| ### `fake/` |
|
|
| Contains **synthetically generated samples** with mismatched acoustic conditions. |
|
|
| --- |
|
|
| ### `protocols/` |
|
|
| Contains dataset splits and evaluation protocols for Our dataset, SceneFake dataset and combined dataset. |
|
|
|
|
|
|
| <file_path> <label> |
|
|
| labels: |
| Real: 0 |
| Fake: 1 |
|
|
| --- |
|
|
|
|
| ## Metadata Files |
|
|
| ### `SceneFake-Wild-Real-Info.csv` |
|
|
| Metadata for **real audio samples**. |
|
|
| | Column | Description | |
| | ---------- | ----------------------------------------------- | |
| | File name | Unique audio identifier | |
| | Speaker Id | Speaker identifier | |
| | Device Id | Recording device identifier | |
| | Langauage | Language of speech (Hindi / English) | |
| | Scene | Acoustic scene label | |
| | Script | Prompt/script used for generating the utterance | |
| | Time (s) | Duration of audio (in seconds) | |
| | Transcript | Final spoken content | |
|
|
|
|
| --- |
|
|
| ### `SceneInfo.csv` |
|
|
| Scene-level information. |
| | Column | Description | |
| | ------------------------- | ----------------------------------------- | |
| | SceneID | Unique identifier for each acoustic scene | |
| | Scene Category | Name/type of the acoustic environment | |
| | # Utterances | Total number of utterances in the scene | |
| | Total time (s) | Total duration of all audio in seconds | |
| | Total time (hours) | Total duration in hours | |
| | # English utterances | Number of English utterances | |
| | Total English time (secs) | Total English audio duration (seconds) | |
| | Total English time (hrs) | Total English audio duration (hours) | |
| | Total Hindi Utterances | Number of Hindi utterances | |
| | Total Hindi time (secs) | Total Hindi audio duration (seconds) | |
| | Total Hindi time (hours) | Total Hindi audio duration (hours) | |
|
|
|
|
|
|
| --- |
|
|
| ## SpeakerInfo.csv |
|
|
| | Column | Description | |
| | ---------------------------- | --------------------------------------- | |
| | speaker_id | Unique speaker identifier | |
| | Age | Age of the speaker | |
| | Gender | Gender of the speaker | |
| | Mother tongue/First language | Native or first language of the speaker | |
| | State | State of origin of the speaker | |
| |
| --- |
| <!-- |
| ## Dataset License and Usage Guidelines |
| This dataset is governed by the **CC BY-NC-ND 4.0** license alongside the specific **Terms of Use** detailed below. |
| |
| |
| ### 1. Intended Use |
| This dataset may be used solely for research and development of audio deepfake detection systems. Any use outside this scope is strictly prohibited. |
| |
| ### 2. Non-Commercial Use |
| Commercial use of the dataset is strictly prohibited. |
| |
| ### 3. Attribution and Citation |
| Any use of the dataset in research or academic work must include proper credit and citation of the repository and its associated paper, including in presentations and publications. |
| |
| ### 4. Misuse and Responsibility |
| Misuse of the voice data may result in legal consequences. Users must ensure the privacy and integrity of the dataset are maintained. |
| Users must not: |
| |
| - misuse the voice data for impersonation, synthesis, or harmful applications |
| - attempt to identify individuals or violate privacy |
| |
| Users are fully responsible for ensuring ethical use and compliance with applicable laws. |
| Misuse may result in legal action. |
| |
| ### 5. Access and Revocation |
| The dataset owner reserves the right to revoke access at any time without prior notice. |
| --> |
| |
| ## Citation |
| |
| If you use this dataset, please cite: |
| |
| ```bibtex |
| @dataset{reality_sf26, |
| author={Sinha, Jiya and S, Aarthi and Agarwal, Akshay and Ratha, Nalini}, |
| title = {Reality-Aware Audio SceneFake Dataset}, |
| year = {2026}, |
| publisher = {Hugging Face}, |
| url = {} |
| } |
