jiaheillu/sovits_audio_preview
收藏Hugging Face2023-04-16 更新2024-03-04 收录
下载链接：
https://hf-mirror.com/datasets/jiaheillu/sovits_audio_preview
下载链接
链接失效反馈官方服务：
资源简介：
---
license: openrail
task_categories:
- conversational
language:
- aa
tags:
- music
size_categories:
- n<1K
pretty_name: genshin_voice_sovits
---
# 预览[.](https://huggingface.co/datasets/jiaheillu/sovits_audio_preview/blob/main/README.md)
**简体中文**｜
[English](https://huggingface.co/datasets/jiaheillu/sovits_audio_preview/blob/main/README_EN.md)｜
[日本語](https://huggingface.co/datasets/jiaheillu/sovits_audio_preview/blob/main/README_JP.md)
本仓库用于预览so-vits-svc-4.0训练出的各种语音模型的效果，**点击角色名**自动跳转对应训练参数。</br>
推荐用**谷歌浏览器**，其他浏览器可能无法正确加载预览的音频。</br>
正常说话的音色转换较为准确，歌曲包含较广的音域且bgm和声等难以去除干净，效果有所折扣。</br>
有推荐的歌想要转换听听效果,或者其他内容建议，[**点我**](https://huggingface.co/datasets/jiaheillu/audio_preview/discussions/new)发起讨论</br>
下面是预览音频，**上下左右滑动**可以看到全部
<style>
.scrolling-container {
width: 100%;
max-width: 1600px;
height: 420px;
overflow: auto;
margin: 0;
}
@media screen and (max-width: 768px) {
.scrolling-container {
width: 100%;
height: 120px
overflow: auto;
}
}
</style>
<div class="scrolling-container">
<table border="1" style="white-space: nowrap; text-align: center;">
<thead>
<tr>
<th>角色名</th>
<th>角色原声A</th>
<th>被转换人声B</th>
<th>A音色替换B</th>
<th>A音色翻唱(点击直接下载)</th>
</tr>
</thead>
<tbody>
<tr>
<td><a href="https://huggingface.co/datasets/jiaheillu/audio_preview/blob/main/散兵效果预览/训练参数速览.md">散兵</a></td>
<td><audio src="https://huggingface.co/datasets/jiaheillu/audio_preview/resolve/main/散兵效果预览/部分训练集/真遗憾，小吉祥草王让他消除了那么多的切片，剥夺了我将他一片一片千刀万剐的快乐%E3%80%82.mp3" controls="controls"></audio></td>
<td><audio src="https://huggingface.co/datasets/jiaheillu/audio_preview/resolve/main/散兵效果预览/原声/shenli3.wav" controls="controls"></audio></td>
<td><audio src="https://huggingface.co/datasets/jiaheillu/audio_preview/resolve/main/散兵效果预览/转换结果/shenli3mp3_auto_liulangzhe.wav" controls="controls"></audio></td>
<td><a href="https://huggingface.co/datasets/jiaheillu/audio_preview/resolve/main/散兵效果预览/转换结果/夢で逢えたら2liulangzhe_f.wav">夢で会えたら</a></td>
</tr>
<tr>
<td><a href="https://huggingface.co/datasets/jiaheillu/audio_preview/blob/main/胡桃_preview/README.md">胡桃</a></td>
<td><audio src="https://huggingface.co/datasets/jiaheillu/audio_preview/resolve/main/%E8%83%A1%E6%A1%83_preview/hutao.wav" controls="controls"></audio></td>
<td>.........</td>
<td>.........</td>
<td>
<a href="https://huggingface.co/datasets/jiaheillu/audio_preview/resolve/main/胡桃_preview/moonlight_shadow2胡桃.WAV">moonlight shadow</a>,
<a href="https://huggingface.co/datasets/jiaheillu/audio_preview/resolve/main/胡桃_preview/云烟成雨2胡桃.WAV">云烟成雨</a>,
<a href="https://huggingface.co/datasets/jiaheillu/audio_preview/resolve/main/胡桃_preview/原点2胡桃.WAV">原点</a>,
<a href="https://huggingface.co/datasets/jiaheillu/audio_preview/resolve/main/胡桃_preview/夢だ会えたら2胡桃.WAV">夢で逢えたら</a>,
<a href="https://huggingface.co/datasets/jiaheillu/audio_preview/resolve/main/胡桃_preview/贝加尔湖畔2胡桃.WAV">贝加尔湖畔</a>
</td>
</tr>
<tr>
<td><a href="https://huggingface.co/datasets/jiaheillu/audio_preview/blob/main/绫华_preview/README.md">神里绫华</a></td>
<td><audio src="https://huggingface.co/datasets/jiaheillu/audio_preview/resolve/main/绫华_preview/linghua428.wav" controls="controls"></audio></td>
<td><audio src="https://huggingface.co/datasets/jiaheillu/audio_preview/resolve/main/绫华_preview/yelan.wav" controls="controls"></audio></td>
<td><audio src="https://huggingface.co/datasets/jiaheillu/sovits_audio_preview/resolve/main/绫华_preview/yelan.wav_auto_linghua_0.5.wav" controls="controls"></audio></td>
<td>
<a href="https://huggingface.co/datasets/jiaheillu/audio_preview/resolve/main/绫华_preview/アムリタ2绫华.WAV">アムリタ</a>,
<a href="https://huggingface.co/datasets/jiaheillu/audio_preview/resolve/main/绫华_preview/大鱼2绫华.WAV">大鱼</a>,
<a href="https://huggingface.co/datasets/jiaheillu/audio_preview/resolve/main/绫华_preview/遊園施設2绫华.WAV">遊園施設</a>,
<a href="https://huggingface.co/datasets/jiaheillu/audio_preview/resolve/main/绫华_preview/the_day_you_want_away2绫华.WAV">the day you want away</a>
</td>
</tr>
<tr>
<td><a href="https://huggingface.co/datasets/jiaheillu/sovits_audio_preview/blob/main/宵宫_preview/README.md">宵宫</a></td>
<td><audio src="https://huggingface.co/datasets/jiaheillu/sovits_audio_preview/resolve/main/宵宫_preview/xiaogong.wav" controls="controls"></audio></td>
<td><audio src="https://huggingface.co/datasets/jiaheillu/sovits_audio_preview/resolve/main/宵宫_preview/hutao2.wav" controls="controls"></audio></td>
<td><audio src="https://huggingface.co/datasets/jiaheillu/sovits_audio_preview/resolve/main/宵宫_preview/hutao2wav_0key_xiaogong_0.5-2.wav" controls="controls"></audio></td>
<td>
<a href="https://huggingface.co/datasets/jiaheillu/sovits_audio_preview/resolve/main/宵宫_preview/昨夜书2宵宫.WAV">昨夜书</a>,
<a href="https://huggingface.co/datasets/jiaheillu/sovits_audio_preview/resolve/main/宵宫_preview/lemon2宵宫.WAV">lemon</a>,
<a href="https://huggingface.co/datasets/jiaheillu/sovits_audio_preview/resolve/main/宵宫_preview/my_heart_will_go_no2宵宫.WAV">my heart will go on</a>,
</td>
</tr>
<tr>
<td><a href="https://huggingface.co/datasets/jiaheillu/sovits_audio_preview/blob/main/刻晴_preview/README.md">刻晴</a></td>
<td><audio src="https://huggingface.co/datasets/jiaheillu/sovits_audio_preview/resolve/main/刻晴_preview/原_keqing2.wav" controls="controls"></audio></td>
<td><audio src="https://huggingface.co/datasets/jiaheillu/sovits_audio_preview/resolve/main/刻晴_preview/待_xiaogong3.wav" controls="controls"></audio></td>
<td><audio src="https://huggingface.co/datasets/jiaheillu/sovits_audio_preview/resolve/main/刻晴_preview/已_xiaogong2keqing.wav" controls="controls"></audio></td>
<td>
<a href="https://huggingface.co/datasets/jiaheillu/sovits_audio_preview/resolve/main/刻晴_preview/嚣张2刻晴.WAV">嚣张</a>,
<a href="https://huggingface.co/datasets/jiaheillu/sovits_audio_preview/resolve/main/刻晴_preview/ファティマ2刻晴.WAV">ファティマ</a>,
<a href="https://huggingface.co/datasets/jiaheillu/sovits_audio_preview/resolve/main/刻晴_preview/hero2刻晴.WAV">hero</a>,
</td>
</tr>
<tr>
<td><a href="https://huggingface.co/datasets/jiaheillu/sovits_audio_preview/blob/main/可莉_preview/README.md">可莉</a></td>
<td><audio src="https://huggingface.co/datasets/jiaheillu/sovits_audio_preview/resolve/main/可莉_preview/原_keli.wav" controls="controls"></audio></td>
<td><audio src="https://huggingface.co/datasets/jiaheillu/sovits_audio_preview/resolve/main/可莉_preview/待_ying.wav" controls="controls"></audio></td>
<td><audio src="https://huggingface.co/datasets/jiaheillu/sovits_audio_preview/resolve/main/可莉_preview/已_ying2keli.wav" controls="controls"></audio></td>
<td>
<a href="https://huggingface.co/datasets/jiaheillu/sovits_audio_preview/resolve/main/可莉_preview/樱花草2可莉.WAV">樱花草</a>,
<a href="https://huggingface.co/datasets/jiaheillu/sovits_audio_preview/resolve/main/可莉_preview/夢をかなえてドラえもん2可莉.WAV">夢をかなえてドラえもん</a>,
<a href="https://huggingface.co/datasets/jiaheillu/sovits_audio_preview/resolve/main/可莉_preview/sun_shine2可莉.WAV">sun_shine</a>,
</td>
</tr>
<tr>
<td><a href="https://huggingface.co/datasets/jiaheillu/sovits_audio_preview/blob/main/鹿野院平藏_preview/README.md">鹿野院平藏</a></td>
<td><audio src="https://huggingface.co/datasets/jiaheillu/sovits_audio_preview/resolve/main/鹿野院平藏_preview/原_pingzang.wav" controls="controls"></audio></td>
<td><audio src="https://huggingface.co/datasets/jiaheillu/sovits_audio_preview/resolve/main/鹿野院平藏_preview/待_shenzi.wav" controls="controls"></audio></td>
<td><audio src="https://huggingface.co/datasets/jiaheillu/sovits_audio_preview/resolve/main/鹿野院平藏_preview/已_shenzi2pingzang.wav" controls="controls"></audio></td>
<td>
<a href="https://huggingface.co/datasets/jiaheillu/sovits_audio_preview/resolve/main/鹿野院平藏_preview/风继续吹2pingng.WAV">风继续吹</a>,
<a href="https://huggingface.co/datasets/jiaheillu/sovits_audio_preview/resolve/main/鹿野院平藏_preview/小さな恋の歌2pingzang.WAV">小さな恋の歌</a>,
<a href="https://huggingface.co/datasets/jiaheillu/sovits_audio_preview/resolve/main/鹿野院平藏_preview/love_yourself2pingzang.WAV">love_yourself</a>,
</td>
</tr>
<tr>
<td><a href="https://huggingface.co/datasets/jiaheillu/sovits_audio_preview/blob/main/imallryt_preview/README.md">imallryt</a></td>
<td><audio src="https://huggingface.co/datasets/jiaheillu/sovits_audio_preview/resolve/main/imallryt_preview/%E5%8E%9F_IVOL_1%20Care_DRY_120_Am_Main_Vocal.wav" controls="controls"></audio></td>
<td><audio src="https://huggingface.co/datasets/jiaheillu/sovits_audio_preview/resolve/main/imallryt_preview/%E5%BE%85_Lead_A%20minor_DRY.wav" controls="controls"></audio></td>
<td><audio src="https://huggingface.co/datasets/jiaheillu/sovits_audio_preview/resolve/main/imallryt_preview/%E5%B7%B2_Lead_A%20minor_DRYwav_0key_imallryt_0.5.wav" controls="controls"></audio></td>
<td>
<a href="https://huggingface.co/datasets/jiaheillu/sovits_audio_preview/resolve/main/imallryt_preview/海阔天空2imallryt.WAV">海阔天空</a>,
</td>
</tr>
<tr>
<td><a href="https://huggingface.co/datasets/jiaheillu/sovits_audio_preview/blob/main/kagami_preview/README.md">kagami</a></td>
<td><audio src="https://huggingface.co/datasets/jiaheillu/sovits_audio_preview/resolve/main/kagami_preview/原_kagami.wav" controls="controls"></audio></td>
<td><audio src="https://huggingface.co/datasets/jiaheillu/sovits_audio_preview/resolve/main/kagami_preview/待_wendi.wav" controls="controls"></audio></td>
<td><audio src="https://huggingface.co/datasets/jiaheillu/sovits_audio_preview/resolve/main/kagami_preview/已_windi2kagami.wav" controls="controls"></audio></td>
<td>
<a href="https://huggingface.co/datasets/jiaheillu/sovits_audio_preview/resolve/main/kagami_preview/えるの侵蝕_Vocals.wav_-4key_kagami_0.5.flac">えるの侵蝕</a>,
</td>
</tr>
</tbody>
</table>
</div>
关键参数：</br>
audio duration:训练集总时长</br>
epoch: 轮数</br>
其余：</br>
batch_size = 一个step训练的片段数<br>
segments = 音频被切分的片段<br>
step=segments*epoch/batch_size,即模型文件后面数字由来<br>
以散兵为例：</br>
损失函数图像：主要看step 与 loss5，比如：<br>
给一个大致的参考，待转换音频都为高音女生，这是较为刁钻的测试：如图，10min纯净人声，
差不多2800epoch（10000step）就已经出结果了，实际使用的是5571epoch(19500step)的文件，被训练音色和原音色相差几
何，请听上方预览音频。正常训练，10min是较为不足的训练集时长。<br>
[点我查看相关文件](https://huggingface.co/datasets/jiaheillu/audio_preview/tree/main)<br>
提供机构：
jiaheillu
原始信息汇总
数据集概述
数据集名称: genshin_voice_sovits
许可证: openrail
任务类别:
- conversational
语言:
- aa
标签:
- music
大小类别:
- n<1K
美观名称: genshin_voice_sovits
数据集内容
本数据集包含多个角色的语音转换预览，每个角色包括以下音频内容：
- 角色原声A
- 被转换人声B
- A音色替换B
- A音色翻唱（可直接下载）
角色列表及其相关音频链接：
-
散兵
-
胡桃
-
神里绫华
-
宵宫
-
刻晴
-
可莉
-
鹿野院平藏
-
imallryt
-
kagami
关键参数
- audio duration: 训练集总时长
- epoch: 轮数
- batch_size: 一个step训练的片段数
- segments: 音频被切分的片段
- step: segments * epoch / batch_size
示例
以散兵为例，其训练参数如下：
- 损失函数图像参考: 链接
- 训练集时长: 10min纯净人声
- 轮数: 5571epoch
- 步数: 19500step
