voices-with-captions

# 🗣️ Synthetic Voice Description Dataset for Voice Conversion

## Overview

This dataset contains **3.180 synthetically generated voice samples**, each paired with a **caption** that roughly describes how the voice sounds. Captions typically include information about the **age**, **gender**, **accent**, and sometimes **general vocal qualities** (e.g., "old woman, Irish accent").

All voices are **synthetically generated** and do not represent any real person. This makes the dataset ideal for training and evaluating **voice conversion models** without infringing on personhood or identity rights.

---

## 📁 Dataset Contents

- A **CSV file** containing pairs of:
- `description`: A short text caption describing the voice
- `matched_filename`: The name of the corresponding `.wav` file

**Example rows from the CSV:**

```csv
description,matched_filename
"old woman, Irish accent",00399-000.wav
"old man, valley girl accent",01074-001.wav
"young man, Scottish accent",18297-001.wav
"middle-aged man, Chinese accent",00952-002.wav
```

- A **TAR archive** that contains:
- The `voice-captions.csv` file (as described above)
- All 3.180 corresponding `.wav` audio files

---

## 🎯 Purpose

The dataset is designed to serve as a **target files** for voice conversion tasks. It can be used to:

- Convert existing voice data into synthetic voices with known, described properties

---

## ⚙️ Google Colab for Seed-VC Fine-Tuning

A community-developed fine-tuning notebook for **Seed-VC** (a voice conversion model) is available on Google Colab:

🔗 **[Seed-VC Fine-Tune Notebook](https://colab.research.google.com/drive/1HeJgMIRpEMd87z5oAcfBfS8_YRLvrwr9?usp=sharing)**

This notebook allows you to take real or synthetic voice data and convert it into one of the target voices.

# 🗣️ 语音转换（Voice Conversion）用合成语音描述数据集

## 概述
本数据集包含**3180个合成生成的语音样本**，每个样本均搭配一段**语音描述标注**，用于大致描述该语音的声学特征。标注内容通常涵盖**年龄**、**性别**、**口音**，有时还会包含**通用嗓音特质**（例如“老年女性，爱尔兰口音”）。

所有语音均为合成生成，不代表任何真实自然人，因此本数据集可安全用于训练与评估**语音转换模型**，不会侵犯他人人格权与身份权益。

---

## 📁 数据集内容
- 一个**CSV文件**，包含以下成对字段：
- `description`：描述该语音的简短文本标注
- `matched_filename`：对应`.wav`音频文件的文件名

**CSV文件示例行：**
csv
description,matched_filename
"老年女性，爱尔兰口音",00399-000.wav
"老年男性，谷区少女口音",01074-001.wav
"年轻男性，苏格兰口音",18297-001.wav
"中年男性，中国口音",00952-002.wav

- 一个**TAR归档文件**，内含：
- 上述的`voice-captions.csv`文件
- 全部3180个对应`.wav`音频文件

---

## 🎯 用途
本数据集旨在作为**语音转换任务的目标语音文件集**，可用于：
- 将现有语音数据转换为具备已知可描述属性的合成语音

---

## ⚙️ 用于Seed-VC微调的Google Colab工具
一款由社区开发的**Seed-VC（语音转换模型）微调笔记本**已发布于Google Colab：
🔗 **[Seed-VC 微调笔记本](https://colab.research.google.com/drive/1HeJgMIRpEMd87z5oAcfBfS8_YRLvrwr9?usp=sharing)**
该笔记本支持将真实或合成语音数据转换为上述目标语音之一。