Add base separator

modules/uvr/music_separator.py

@@ -0,0 +1,48 @@
+# Credit to Team UVR : https://github.com/Anjok07/ultimatevocalremovergui
+from typing import Optional
+import soundfile as sf
+import os
+import torch
+
+from uvr.models import MDX, Demucs, VrNetwork, MDXC
+
+
+class MusicSeparator:
+    def __init__(self,
+                 model_dir: Optional[str] = None,
+                 output_dir: Optional[str] = None):
+        self.model = None
+        self.device = self.get_device()
+        self.model_dir = model_dir
+        self.output_dir = output_dir
+
+    def update_model(self,
+                     model_name: str = "UVR-MDX-NET-Inst_1",
+                     segment_size: int = 256):
+        self.model = MDX(name="UVR-MDX-NET-Inst_1",
+                         other_metadata={"segment": segment_size, "split": True},
+                         device=self.device,
+                         logger=None,
+                         model_path="models\UVR\MDX_Net_Models\UVR-MDX-NET-Inst_HQ_1.onnx")
+
+    def separate(self,
+                 audio_file_path: str,
+                 sample_rate: int = 44100):
+        if self.model is None:
+            self.model = self.update_model()
+
+        filename = audio_file_path
+        instrumental_output_path = os.path.join(self.output_dir, "instrumental", filename)
+        vocals_output_path = os.path.join(self.output_dir, "vocals", filename)
+
+        result = self.model(audio_file_path)
+        instrumental, vocals = result["instrumental"], result["vocals"]
+
+        sf.write('instrumental.wav', instrumental.T, sample_rate, format="WAV")
+        sf.write('vocals.wav', vocals.T, sample_rate, format="WAV")
+
+        return instrumental_output_path, vocals_output_path
+
+    @staticmethod
+    def get_device():
+        return "cuda" if torch.cuda.is_available() else "cpu"