Youka 内置了多种 AI 模型，用于将人声与伴奏分离。每个模型都在质量、速度以及它在伴奏轨中保留的内容之间有所取舍。Documentation Index
Fetch the complete documentation index at: https://docs.youka.io/llms.txt
Use this file to discover all available pages before exploring further.
快速对比
| 模型 | 输出分轨 | 最适合 |
|---|---|---|
| MDX-23C (default) | Instrumental, Lead vocals | 大多数卡拉 OK 项目 |
| AudioShakeAI | Instrumental, Lead vocals | 最高质量的分离 |
| MusicAI (Instrumental Only) | Instrumental | 只需要伴奏轨时 |
| MusicAI (Instrumental / Backing / Lead) | Instrumental, Backing vocals, Lead vocals | 需要对和声进行完整混音控制 |
| MusicAI (With Backing Vocals) | Instrumental + backing, Lead vocals | 希望合唱/和声保持可听的卡拉 OK |
| MusicAI (Without Backing Vocals) | Instrumental, Lead vocals | 纯净伴奏、无和声 |
| MDX-Net (With Backing Vocals) | Instrumental + backing, Lead vocals | 作为 MusicAI backing-vocal model 的替代方案 |
| RoFormer (and variants) | Instrumental, Lead vocals | 实验性 — 如果默认模型不适合这首歌可以尝试 |
| Demucs | Instrumental, Lead vocals | 开源备选方案 |
创建卡拉 OK 时选择模型
在现有项目上尝试不同模型
相关
- 保留背景人声：适用于合唱与和声
- 在特定行保留原始人声