dolly-vn/dolly-audio-1000h-vietnamese

---
configs:
- config_name: default
data_files:
- split: train
path: data/train-*
dataset_info:
features:
- name: audio_filename
dtype: string
- name: text
dtype: string
- name: voice_id
dtype: string
- name: audio
dtype:
audio:
decode: false
splits:
- name: train
num_bytes: 165955597080
num_examples: 664125
download_size: 157800059320
dataset_size: 165955597080
language:
- vi
tags:
- vietnamese
- synthetic
- audio
- tts
size_categories:
- 100K<n<1M
---
# *Dolly-Audio: Vietnamese Multi-Speaker High-Quality Speech Corpus*

## *Dataset Summary*

*Dolly-Audio* is a large-scale, high-quality Vietnamese speech corpus created by the *Dolly AI Team*.
Inspired by Dolly, the world’s first cloned mammal, the project aims to advance research in Vietnamese speech synthesis, speech recognition, and voice modeling.

This release provides nearly *1,000 hours of professionally cleaned audio*, featuring *152 speakers* across different Vietnamese regions and speaking styles. Text transcripts span a wide variety of domains to ensure linguistic diversity and model robustness.

---

## *Key Features*

* ~1,000 hours of high-quality Vietnamese speech
* 152 multi-region speakers with diverse accents
* Cleaned, noise-free audio; no background music
* Sentence-level boundary trimming for natural prosody
* Rich transcript domains (news, entertainment, education, conversational, etc.)
* Estimated *near-zero WER (≈ 0%)* from manual sampling
* Suitable for TTS, ASR, voice cloning, and speech research

---

## *Intended Use*

The dataset is ideal for:

* Multi-speaker text-to-speech (TTS)
* Automatic speech recognition (ASR)
* Voice cloning and speaker adaptation
* Prosody modeling
* Linguistic and phonetic research

Commercial use is *not permitted* under the license.

---

## *Usage Restrictions*

* Non-commercial *research use only*
* Redistribution must comply with *CC-BY-NC-SA-4.0*
* Users must verify dataset suitability for their research task
* Institutional email required for access approval

---

## *Citation*

If you use Dolly-Audio in your research, please credit the creators:

*Nguyen Vinh Huy* — [[e-mail]](mailto:[e-mail])
*Nguyen Dinh Thuan* — [[e-mail]](mailto:[e-mail])

@dataset{dolly_audio_2025,
title = {Dolly-Audio: Vietnamese Multi-Speaker High-Quality Speech Corpus},
author = {Nguyen, Vinh Huy and Nguyen, Dinh Thuan},
year = {2025},
publisher = {Dolly AI Team},
howpublished = {\url{https://huggingface.co/datasets/Dolly-AI/Dolly-Audio}},
note = {Released under CC-BY-NC-SA-4.0. Research use only.}
}

---

## *Contact*

For access requests or inquiries, please contact the maintainers via the emails above.

configs:
- config_name: default
data_files:
- split: train
path: data/train-*
dataset_info:
features:
- name: audio_filename
dtype: string
- name: text
dtype: string
- name: voice_id
dtype: string
- name: audio
dtype:
audio:
decode: false
splits:
- name: train
num_bytes: 165955597080
num_examples: 664125
download_size: 157800059320
dataset_size: 165955597080
language:
- vi
tags:
- vietnamese
- synthetic
- audio
- tts
size_categories:
- 100K<n<1M

# Dolly-Audio：越南语多说话人高质量语音语料库

## 数据集概览

*Dolly-Audio* 是由Dolly AI团队（Dolly AI Team）打造的大规模高质量越南语语音语料库。本项目得名于世界首只克隆哺乳动物多莉，旨在推动越南语语音合成、语音识别以及语音建模领域的研究。

本次发布的语料库包含近1000小时经过专业清洗的音频，涵盖来自越南不同地区、具备多样说话风格的152名说话人。文本转录内容覆盖多领域，以保障语言多样性与模型鲁棒性。

---

## 核心特性

* 近1000小时高质量越南语语音
* 152名多地域说话人，口音丰富多样
* 经过清洗的无噪音频，无背景音乐干扰
* 句子级边界裁切，实现自然韵律
* 丰富的转录领域（新闻、娱乐、教育、对话等）
* 经人工抽样估算，词错误率（Word Error Rate, WER）近乎为0（≈0%）
* 适用于文本转语音（TTS）、自动语音识别（ASR）、语音克隆及语音研究领域

---

## 适用场景

本数据集适用于：
* 多说话人文本转语音（TTS）
* 自动语音识别（ASR）
* 语音克隆与说话人自适应
* 韵律建模
* 语言学与语音学研究

根据授权协议，商业使用是不被允许的。

---

## 使用限制

* 仅可用于非商业研究用途
* 再分发需遵守CC-BY-NC-SA-4.0协议
* 用户需自行验证数据集是否适配自身研究任务
* 申请访问需提供机构邮箱

---

## 引用方式

若您在研究中使用Dolly-Audio，请引用如下创作者：
* 阮文惠（Nguyen Vinh Huy）—— [[e-mail]](mailto:[e-mail])
* 阮丁顺（Nguyen Dinh Thuan）—— [[e-mail]](mailto:[e-mail])

bibtex
@dataset{dolly_audio_2025,
title = {Dolly-Audio: Vietnamese Multi-Speaker High-Quality Speech Corpus},
author = {Nguyen, Vinh Huy and Nguyen, Dinh Thuan},
year = {2025},
publisher = {Dolly AI Team},
howpublished = {url{https://huggingface.co/datasets/Dolly-AI/Dolly-Audio}},
note = {Released under CC-BY-NC-SA-4.0. Research use only.}
}

---

## 联系方式

如需申请访问或咨询，请通过上述邮箱联系维护人员。