---
tags:
  - audio
  - speech
  - whisper
  - chinese
  - audiovisual
  - parallel
license: cc-by-nc-sa-4.0
extra_gated_heading: "Data Application"
extra_gated_description: |
  Please fill out the form below to apply for access to this dataset.  
  Access will be granted only for academic and non-commercial research use.  
  Make sure to use your institutional email address and agree to the dataset license terms.
extra_gated_button_content: "Submit Application"

extra_gated_fields:
  First Name:
    type: text
    required: true
  Last Name:
    type: text
    required: true
  Affiliation (Research organization):
    type: text
    required: true
  Email Address (Please use institutional e-mail):
    type: text
    required: true
  I have read and agree to the license terms (CC BY-NC-SA 4.0):
    type: checkbox
    required: true
---


# 🗣️ AISHELL6-Whisper

**AISHELL6-Whisper** is a **large-scale open-source Chinese Mandarin audio-visual whisper speech dataset**,  
containing **30 hours each** of whisper and parallel normal speech, with synchronized frontal RGB-D facial videos.

---

## 📘 Dataset Summary

| **Property** | **Description** |
|:-------------|:----------------|
| **Language** | Chinese (Mandarin, ZH) |
| **License** | [CC BY-NC-SA 4.0](https://creativecommons.org/licenses/by-nc-sa/4.0/) |
| **Duration** | ~60 hours total (30 h whisper + 30 h normal) |
| **Speakers** | 167 total (121 with RGB-D, 46 audio-only) |
| **Environment** | Controlled studio recording |
| **Sampling Rate** | 48 kHz |
| **Video Resolution** | 1280 × 720 @ 25 fps |

---

## 🎙️ Dataset Description

### 🏛️ Recording Environment

The **AISHELL6-Whisper** corpus was collected in a **controlled studio** to ensure acoustic consistency between whisper and normal speech.

- **Speakers:** 167 participants reading 10–20 minutes of *unique poetry texts* (no content overlap).  
- **Modalities:**  
  - **121 speakers:** high-fidelity audio + RGB-D video  
  - **46 speakers:** audio only  
- **Speech styles:** Whispered and normal speech, recorded in parallel sessions.

---

### 🔊 Audio Specifications

- Microphone: **Neumann U87** single-channel condenser mic  
- Sampling rate: **48,000 Hz**  
- Background noise level: **< 20 dB (studio noise floor)**  
- File format: `.wav`  
- Each recording contains **paired whisper–normal utterances**

---

### 🎥 Video Specifications

- Camera: **RGB-D** depth camera  
- Placement: **1 meter in front of the speaker**  
- Resolution: **1280×720 pixels**  
- Frame rate: **25 fps**  
- Includes **frontal facial views** synchronized with the audio.

---

## 📄 Dataset Resources

- 📄 **Paper:** [arXiv:2509.23833](https://arxiv.org/abs/2509.23833)  
- 🌐 **Demo:** [zutm.github.io/AISHELL6-Whisper](https://zutm.github.io/AISHELL6-Whisper/)  

---

## 📜 Citation

If you use this dataset, please cite:

```bibtex
@article{li2025aishell6,
  title   = {AISHELL6-Whisper: A Chinese Mandarin Audio-Visual Whisper Speech Dataset with Speech Recognition Baselines},
  author  = {Li, Cancan and Su, Fei and Liu, Juan and Bu, Hui and Wan, Yulong and Suo, Hongbin and Li, Ming},
  journal = {arXiv preprint arXiv:2509.23833},
  year    = {2025}
}