mirror of
https://github.com/aiclientproxy/proxycast.git
synced 2026-10-24 12:04:22 +08:00
368 lines
9.9 KiB
Python
368 lines
9.9 KiB
Python
#!/usr/bin/env python3
|
|
"""
|
|
MP3 转 MIDI 工具
|
|
使用 Demucs 分离人声 + Basic Pitch 转换 MIDI
|
|
|
|
用法:
|
|
python audio_to_midi.py <input_mp3> [output_dir]
|
|
python audio_to_midi.py --check # 检查依赖和硬件
|
|
|
|
输出:
|
|
JSON 格式的处理结果
|
|
"""
|
|
|
|
import sys
|
|
import os
|
|
import json
|
|
import subprocess
|
|
import shutil
|
|
from pathlib import Path
|
|
from datetime import datetime
|
|
|
|
|
|
def output_json(data):
|
|
"""输出 JSON 格式结果"""
|
|
print(json.dumps(data, ensure_ascii=False, indent=2))
|
|
|
|
|
|
def detect_hardware():
|
|
"""检测可用硬件加速"""
|
|
try:
|
|
import torch
|
|
if torch.cuda.is_available():
|
|
device_name = torch.cuda.get_device_name(0)
|
|
return {
|
|
"device": "cuda",
|
|
"name": device_name,
|
|
"description": f"NVIDIA GPU 加速 ({device_name})",
|
|
"estimated_time": "1-2 分钟"
|
|
}
|
|
elif hasattr(torch.backends, 'mps') and torch.backends.mps.is_available():
|
|
return {
|
|
"device": "mps",
|
|
"name": "Apple Silicon",
|
|
"description": "Apple Silicon 加速 (MPS)",
|
|
"estimated_time": "2-3 分钟"
|
|
}
|
|
except ImportError:
|
|
pass
|
|
|
|
return {
|
|
"device": "cpu",
|
|
"name": "CPU",
|
|
"description": "CPU 模式 (较慢)",
|
|
"estimated_time": "8-15 分钟"
|
|
}
|
|
|
|
|
|
def check_dependencies():
|
|
"""检查依赖是否安装"""
|
|
dependencies = {
|
|
"demucs": {"installed": False, "version": None},
|
|
"basic_pitch": {"installed": False, "version": None},
|
|
"torch": {"installed": False, "version": None},
|
|
}
|
|
|
|
try:
|
|
import demucs
|
|
dependencies["demucs"]["installed"] = True
|
|
dependencies["demucs"]["version"] = getattr(demucs, '__version__', 'unknown')
|
|
except ImportError:
|
|
pass
|
|
|
|
try:
|
|
import basic_pitch
|
|
dependencies["basic_pitch"]["installed"] = True
|
|
dependencies["basic_pitch"]["version"] = getattr(basic_pitch, '__version__', 'unknown')
|
|
except ImportError:
|
|
pass
|
|
|
|
try:
|
|
import torch
|
|
dependencies["torch"]["installed"] = True
|
|
dependencies["torch"]["version"] = torch.__version__
|
|
except ImportError:
|
|
pass
|
|
|
|
return dependencies
|
|
|
|
|
|
def check_command_available(cmd):
|
|
"""检查命令行工具是否可用"""
|
|
return shutil.which(cmd) is not None
|
|
|
|
|
|
def separate_vocals(input_mp3, output_dir, device="cpu"):
|
|
"""
|
|
使用 Demucs 分离人声
|
|
|
|
Args:
|
|
input_mp3: 输入 MP3 文件路径
|
|
output_dir: 输出目录
|
|
device: 使用的设备 (cuda/mps/cpu)
|
|
|
|
Returns:
|
|
vocals_path: 人声文件路径
|
|
"""
|
|
input_path = Path(input_mp3)
|
|
output_path = Path(output_dir)
|
|
|
|
# 构建 demucs 命令
|
|
cmd = [
|
|
sys.executable, "-m", "demucs",
|
|
"--two-stems=vocals", # 只分离人声和伴奏
|
|
"-o", str(output_path),
|
|
"--device", device if device != "mps" else "mps",
|
|
]
|
|
|
|
# 添加输入文件
|
|
cmd.append(str(input_path))
|
|
|
|
# 执行命令
|
|
try:
|
|
result = subprocess.run(
|
|
cmd,
|
|
capture_output=True,
|
|
text=True,
|
|
timeout=1800 # 30 分钟超时
|
|
)
|
|
|
|
if result.returncode != 0:
|
|
return None, f"Demucs 执行失败: {result.stderr}"
|
|
|
|
# 查找输出的人声文件
|
|
# Demucs 输出格式: output_dir/htdemucs/song_name/vocals.wav
|
|
song_name = input_path.stem
|
|
vocals_path = output_path / "htdemucs" / song_name / "vocals.wav"
|
|
|
|
if not vocals_path.exists():
|
|
# 尝试其他可能的路径
|
|
for model_dir in output_path.iterdir():
|
|
if model_dir.is_dir():
|
|
possible_path = model_dir / song_name / "vocals.wav"
|
|
if possible_path.exists():
|
|
vocals_path = possible_path
|
|
break
|
|
|
|
if vocals_path.exists():
|
|
return str(vocals_path), None
|
|
else:
|
|
return None, f"未找到人声文件，请检查 {output_path} 目录"
|
|
|
|
except subprocess.TimeoutExpired:
|
|
return None, "Demucs 处理超时 (超过 30 分钟)"
|
|
except Exception as e:
|
|
return None, f"Demucs 执行异常: {str(e)}"
|
|
|
|
|
|
def convert_to_midi(vocals_wav, output_dir):
|
|
"""
|
|
使用 Basic Pitch 将人声转换为 MIDI
|
|
|
|
Args:
|
|
vocals_wav: 人声 WAV 文件路径
|
|
output_dir: 输出目录
|
|
|
|
Returns:
|
|
midi_path: MIDI 文件路径
|
|
"""
|
|
vocals_path = Path(vocals_wav)
|
|
output_path = Path(output_dir)
|
|
|
|
# 构建 basic-pitch 命令
|
|
cmd = [
|
|
sys.executable, "-m", "basic_pitch",
|
|
str(output_path),
|
|
str(vocals_path)
|
|
]
|
|
|
|
try:
|
|
result = subprocess.run(
|
|
cmd,
|
|
capture_output=True,
|
|
text=True,
|
|
timeout=300 # 5 分钟超时
|
|
)
|
|
|
|
if result.returncode != 0:
|
|
return None, f"Basic Pitch 执行失败: {result.stderr}"
|
|
|
|
# 查找输出的 MIDI 文件
|
|
# Basic Pitch 输出格式: output_dir/vocals_basic_pitch.mid
|
|
midi_name = vocals_path.stem + "_basic_pitch.mid"
|
|
midi_path = output_path / midi_name
|
|
|
|
if midi_path.exists():
|
|
return str(midi_path), None
|
|
else:
|
|
# 尝试查找任何 .mid 文件
|
|
for f in output_path.glob("*.mid"):
|
|
return str(f), None
|
|
return None, f"未找到 MIDI 文件，请检查 {output_path} 目录"
|
|
|
|
except subprocess.TimeoutExpired:
|
|
return None, "Basic Pitch 处理超时 (超过 5 分钟)"
|
|
except Exception as e:
|
|
return None, f"Basic Pitch 执行异常: {str(e)}"
|
|
|
|
|
|
def process_audio(input_mp3, output_dir=None):
|
|
"""
|
|
完整的音频处理流程
|
|
|
|
Args:
|
|
input_mp3: 输入 MP3 文件路径
|
|
output_dir: 输出目录 (默认为输入文件所在目录)
|
|
|
|
Returns:
|
|
处理结果字典
|
|
"""
|
|
input_path = Path(input_mp3)
|
|
|
|
if not input_path.exists():
|
|
return {
|
|
"status": "error",
|
|
"error": f"输入文件不存在: {input_mp3}"
|
|
}
|
|
|
|
if output_dir is None:
|
|
output_dir = input_path.parent
|
|
|
|
output_path = Path(output_dir)
|
|
output_path.mkdir(parents=True, exist_ok=True)
|
|
|
|
# 检测硬件
|
|
hardware = detect_hardware()
|
|
|
|
# 检查依赖
|
|
deps = check_dependencies()
|
|
missing_deps = [name for name, info in deps.items()
|
|
if not info["installed"] and name != "torch"]
|
|
|
|
if missing_deps:
|
|
return {
|
|
"status": "error",
|
|
"error": "缺少必要依赖",
|
|
"missing_dependencies": missing_deps,
|
|
"install_command": f"pip install {' '.join(missing_deps).replace('_', '-')}",
|
|
"alternative": "或使用在线工具: https://basicpitch.spotify.com"
|
|
}
|
|
|
|
result = {
|
|
"status": "processing",
|
|
"input_file": str(input_path),
|
|
"output_dir": str(output_path),
|
|
"hardware": hardware,
|
|
"steps": []
|
|
}
|
|
|
|
# Step 1: 分离人声
|
|
result["steps"].append({
|
|
"step": 1,
|
|
"name": "分离人声",
|
|
"status": "in_progress",
|
|
"tool": "Demucs"
|
|
})
|
|
|
|
vocals_path, error = separate_vocals(
|
|
input_mp3,
|
|
output_path,
|
|
hardware["device"]
|
|
)
|
|
|
|
if error:
|
|
result["status"] = "error"
|
|
result["steps"][-1]["status"] = "failed"
|
|
result["steps"][-1]["error"] = error
|
|
return result
|
|
|
|
result["steps"][-1]["status"] = "completed"
|
|
result["steps"][-1]["output"] = vocals_path
|
|
result["vocals_file"] = vocals_path
|
|
|
|
# Step 2: 转换为 MIDI
|
|
result["steps"].append({
|
|
"step": 2,
|
|
"name": "转换 MIDI",
|
|
"status": "in_progress",
|
|
"tool": "Basic Pitch"
|
|
})
|
|
|
|
midi_path, error = convert_to_midi(vocals_path, output_path)
|
|
|
|
if error:
|
|
result["status"] = "error"
|
|
result["steps"][-1]["status"] = "failed"
|
|
result["steps"][-1]["error"] = error
|
|
return result
|
|
|
|
result["steps"][-1]["status"] = "completed"
|
|
result["steps"][-1]["output"] = midi_path
|
|
result["midi_file"] = midi_path
|
|
|
|
# 重命名 MIDI 文件为更友好的名称
|
|
final_midi_name = input_path.stem + ".mid"
|
|
final_midi_path = output_path / final_midi_name
|
|
|
|
if str(midi_path) != str(final_midi_path):
|
|
try:
|
|
shutil.move(midi_path, final_midi_path)
|
|
result["midi_file"] = str(final_midi_path)
|
|
except Exception:
|
|
pass # 保持原文件名
|
|
|
|
result["status"] = "success"
|
|
result["message"] = "MP3 转 MIDI 完成"
|
|
result["completed_at"] = datetime.now().isoformat()
|
|
|
|
return result
|
|
|
|
|
|
def main():
|
|
"""主函数"""
|
|
if len(sys.argv) < 2:
|
|
output_json({
|
|
"status": "error",
|
|
"error": "缺少参数",
|
|
"usage": "python audio_to_midi.py <input_mp3> [output_dir]",
|
|
"examples": [
|
|
"python audio_to_midi.py song.mp3",
|
|
"python audio_to_midi.py song.mp3 ./output",
|
|
"python audio_to_midi.py --check"
|
|
]
|
|
})
|
|
sys.exit(1)
|
|
|
|
# 检查模式
|
|
if sys.argv[1] == "--check":
|
|
deps = check_dependencies()
|
|
hardware = detect_hardware()
|
|
|
|
all_installed = all(
|
|
info["installed"]
|
|
for name, info in deps.items()
|
|
if name != "torch"
|
|
)
|
|
|
|
output_json({
|
|
"status": "ready" if all_installed else "missing_dependencies",
|
|
"dependencies": deps,
|
|
"hardware": hardware,
|
|
"install_command": "pip install demucs basic-pitch" if not all_installed else None,
|
|
"online_alternative": "https://basicpitch.spotify.com"
|
|
})
|
|
sys.exit(0 if all_installed else 1)
|
|
|
|
# 处理模式
|
|
input_mp3 = sys.argv[1]
|
|
output_dir = sys.argv[2] if len(sys.argv) > 2 else None
|
|
|
|
result = process_audio(input_mp3, output_dir)
|
|
output_json(result)
|
|
|
|
sys.exit(0 if result["status"] == "success" else 1)
|
|
|
|
|
|
if __name__ == "__main__":
|
|
main()
|