--- license: gpl-3.0 task_categories: - audio-to-audio language: - zh tags: - rvc - dataset - speech - mozi pretty_name: Mozi Legacy Voice Dataset (15yo Ver.) --- # 📼 Mozi Legacy Voice Dataset (15yo Ver.) ![License: GPL v3](https://img.shields.io/badge/License-GPLv3-blue.svg) ![Status: Archived](https://img.shields.io/badge/Status-Archived%20(Legacy)-orange) ## 📖 简介 / Introduction 这是一个用于训练 RVC (Retrieval-based Voice Conversion) 模型的原始音频数据集。 声音来源是 **15岁时的 Mozi**。 这份数据是从旧电脑的硬盘深处挖掘出来的“数字遗物”。由于作者本人(Mozi)已经经历了变声期,生理结构发生了不可逆的变化(声带变厚、共鸣腔改变),目前的 Mozi-3/4 模型已无法复刻此版本的音色。因此,这份数据在物理层面上是**绝版**的。 This is the raw audio dataset used to train the legacy Mozi RVC model. The voice belongs to **Mozi at the age of 15**. Since the author's voice has naturally deepened and changed over time, this specific timbre is biologically impossible to reproduce today. This makes the dataset a "time capsule" of a voice that no longer exists in the physical world. ## 🎤 数据集特点 / Features * **少年音色 (Juvenile Tone)**: 包含了变声期特有的不稳定性与青涩感。 * **高动态范围 (High Dynamic Range)**: 数据集中不仅包含正常的念白/演唱,还包含了一些**高强度的情绪表达**(来源于当时游玩恐怖游戏时的真实尖叫录音)。 * **Lo-Fi 质感**: 由于录制于早期,部分音频可能带有当年的设备底噪,这构成了一种独特的 Lo-Fi 听感特征。 ## 🛠️ 用法 / Usage 本数据集适用于: 1. **RVC 模型训练**: 复刻那个“传说中”的 Mozi Legacy 模型。 2. **鬼畜/二创素材**: 其中的尖叫和情绪化音频非常适合作为 VOCALOID 风格或 meme 视频的素材。 3. **声音合成研究**: 研究变声期少年的声学特征。 ### ⚠️ 注意事项 / Note 由于包含部分高分贝尖叫,请在试听或处理数据时注意调整耳机音量。 *Warning: Contains loud screams. Please adjust your volume.* ## 📜 协议 / License 本数据集遵循 **GNU General Public License v3.0 (GPL-3.0)** 协议。 这意味着: * ✅ 你可以自由地使用、修改、分发此数据集。 * ✅ 你可以使用此数据集训练新的 AI 模型。 * 🔄 **Copyleft**: 如果你发布了基于此数据集修改后的版本,或者发布了基于此数据集训练的模型文件,你必须在相同的 GPL 3.0 协议下开源你的成果。 ## 🔗 关于作者 / About Author * **Web**: [mozi1924.com](https://mozi1924.com) --- *Created with ❤️ by Mozi (and the ghosts of the past).* ``` ---