docs: update demucs-separate — add separate_full MIDI replace mode

This commit is contained in:
yumoqing 2026-07-06 11:55:45 +08:00
parent de3fcfa8a9
commit 3eb7d5514c

View File

@ -1886,20 +1886,54 @@ curl -X POST 'https://token.opencomputing.cn/llmage/v1/pipeline/submit' \
### 2. 音频分离 — ky-demucs-separate
分离音频中的人声和伴奏。
分离音频中的人声和伴奏。支持两种模式。
**分离模式:**
| mode | `task_type` | 说明 |
|------|-------------|------|
| 两轨分离(默认) | `separate` | 分离人声(vocals)和伴奏(no_vocals)返回2个文件 |
| 全轨MIDI替换 | `separate_full` | 分离人声+drums+bass+other四轨每个乐器轨转MIDI后用GM音源替换混音为伴奏 |
**全轨MIDI替换流程**
1. demucs 4轨分离 → vocals / drums / bass / other
2. 每个乐器轨 basic-pitch → MIDI
3. MIDI 用 FluidR3_GM SoundFont 渲染drums=ch10鼓组, bass=prog34电贝司, other=prog1钢琴
4. ffmpeg 混音 → accompaniment.wav
| 参数 | 类型 | 必填 | 说明 |
|------|------|------|------|
| `audio_url` | string | 是 | 音频文件URL |
| `task_type` | string | 否 | `separate`(默认,两轨)或 `separate_full`全轨MIDI替换 |
```bash
# 两轨分离(默认)
curl -X POST 'https://token.opencomputing.cn/llmage/v1/pipeline/submit' \
-H 'Authorization: Bearer ***' \
-H 'Content-Type: application/json' \
-d '{"model":"ky-demucs-separate","audio_url":"https://example.com/music.mp3"}'
# 全轨MIDI替换
curl -X POST 'https://token.opencomputing.cn/llmage/v1/pipeline/submit' \
-H 'Authorization: Bearer ***' \
-H 'Content-Type: application/json' \
-d '{"model":"ky-demucs-separate","audio_url":"https://example.com/music.mp3","task_type":"separate_full"}'
```
**响应:**
**响应(两轨模式):**
```json
{
"taskid": "luid_xxx",
"taskstatus": "SUCCEEDED",
"vocals_url": "https://token.opencomputing.cn/.../vocals.wav",
"accompaniment_url": "https://token.opencomputing.cn/.../no_vocals.wav",
"mode": "2stem",
"usage": {"audio_seconds": 240.0}
}
```
**响应全轨MIDI替换模式**
```json
{
@ -1907,6 +1941,10 @@ curl -X POST 'https://token.opencomputing.cn/llmage/v1/pipeline/submit' \
"taskstatus": "SUCCEEDED",
"vocals_url": "https://token.opencomputing.cn/.../vocals.wav",
"accompaniment_url": "https://token.opencomputing.cn/.../accompaniment.wav",
"drums_midi_url": "https://token.opencomputing.cn/.../drums.wav",
"bass_midi_url": "https://token.opencomputing.cn/.../bass.wav",
"other_midi_url": "https://token.opencomputing.cn/.../other.wav",
"mode": "full",
"usage": {"audio_seconds": 240.0}
}
```