---
task_categories:
- audio-to-audio
- text-to-speech
- audio-classification
language:
- en
---
# NaturalVoices EVC

A large emotional voice conversion (EVC) dataset curated from spontaneous, in-the-wild podcast speech as part of the **NaturalVoices** project in collaboration with 🤗[MSP Lab at CMU LTI](https://huggingface.co/Lab-MSP). This release provides the emotion balanced subset of the NaturalVoices **870-hour** VC dataset and intended for training and evaluating emotion-aware voice conversion systems but not limited to VC tasks.

- 📄 Paper: *NaturalVoices: A Large-Scale, Spontaneous and Emotional Podcast Dataset for Voice Conversion* — https://arxiv.org/abs/2511.00256 \
- 🧺 Dataset collection (related subsets, e.g., 10% of data & emotional VC): https://huggingface.co/collections/JHU-SmileLab/naturalvoices-voice-conversion-datasets \
- <span style="display:inline-flex;align-items:center;gap:-6px">
  <img src="https://img.shields.io/badge/github-%23121011.svg?style=for-the-badge&logo=github&logoColor=white" height=20 alt="GitHub badge">
  <span>The extensive (unfiltered) NaturalVoices dataset and the code for the data collection & curation pipeline: <a href="https://github.com/Lab-MSP/NaturalVoices">https://github.com/Lab-MSP/NaturalVoices</a></span>

</span>

## Dataset Summary

NaturalVoices VC compiles real-life, expressive podcast speech and provides automatic **annotations** designed for VC research (e.g., **emotion** attributes, **speaker identity**, **speech quality**, **transcripts**). The broader NaturalVoices corpus contains thousands of hours of podcast speech; this repository hosts the **EVC** subset.

**What’s in this repo**

- ~370 hours of podcast speech tailored and preprocessed for EVC.
- Balanced distribution of categorical emotions (Angry, Happy, Neutral, Sad)
- A wide range of speakers both manually & automatically annotated.
- Annotations archive with per-utterance annotations including:

  - Emotion categorical labels & dimensional attributes (valence/arousal/dominance),
  - Speech quality indicators,
  - Text, Gender, and Duration.

### Subsets
| Subset                    | Description                                                  | Link                                                                                                     |
| --------------------------- | :------------------------------------------------------------- | ---------------------------------------------------------------------------------------------------------- |
| NaturalVoices_VC_870h     | 870h of speech data curated for VC                           | 🤗[JHU-SmileLab/NaturalVoices_VC_870h](https://JHU-SmileLab/NaturalVoices_VC_870h)                       |
| NaturalVoices_EVC         | Emotion-balanced subset for Emotional Voice Conversion (EVC) | This repo                                                                                                |
| NaturalVoices_VC_01 (10%) | A smaller subset uniformly sampled from 870h (10%)           | 🤗[JHU-SmileLab/NaturalVoices_VC_0.1](https://huggingface.co/datasets/JHU-SmileLab/NaturalVoices_VC_0.1) |

## How to use

You can directly download the dataset using the following command:

```bash
huggingface-cli download JHU-SmileLab/NaturalVoices_EVC --repo-type=dataset --local-dir=YOUR_LOCAL_DIR 
```

*Streaming support will be available*

## Cite & Contribute

If you use this dataset, please cite the paper:

```sql
@misc{du2025naturalvoiceslargescalespontaneousemotional,
      title={NaturalVoices: A Large-Scale, Spontaneous and Emotional Podcast Dataset for Voice Conversion}, 
      author={Zongyang Du and Shreeram Suresh Chandra and Ismail Rasim Ulgen and Aurosweta Mahapatra and Ali N. Salman and Carlos Busso and Berrak Sisman},
      year={2025},
      eprint={2511.00256},
      archivePrefix={arXiv},
      primaryClass={eess.AS},
      url={https://arxiv.org/abs/2511.00256}, 
}
```