Upload f0.py

f0.py

@@ -0,0 +1,84 @@
+import os
+import torch
+
+from rvc.lib.predictors.RMVPE import RMVPE0Predictor
+from torchfcpe import spawn_bundled_infer_model
+import torchcrepe
+
+
+class RMVPE:
+    def __init__(self, device, model_name="rmvpe.pt", sample_rate=16000, hop_size=160):
+        self.device = device
+        self.sample_rate = sample_rate
+        self.hop_size = hop_size
+        self.model = RMVPE0Predictor(
+            os.path.join("rvc", "models", "predictors", model_name),
+            device=self.device,
+        )
+
+    def get_f0(self, x, filter_radius=0.03):
+        f0 = self.model.infer_from_audio(x, thred=filter_radius)
+        return f0
+
+
+class CREPE:
+    def __init__(self, device, sample_rate=16000, hop_size=160):
+        self.device = device
+        self.sample_rate = sample_rate
+        self.hop_size = hop_size
+
+    def get_f0(self, x, f0_min=50, f0_max=1100, p_len=None, model="full"):
+        if p_len is None:
+            p_len = x.shape[0] // self.hop_size
+
+        if not torch.is_tensor(x):
+            x = torch.from_numpy(x)
+
+        batch_size = 512
+
+        f0, pd = torchcrepe.predict(
+            x.float().to(self.device).unsqueeze(dim=0),
+            self.sample_rate,
+            self.hop_size,
+            f0_min,
+            f0_max,
+            model=model,
+            batch_size=batch_size,
+            device=self.device,
+            return_periodicity=True,
+        )
+        pd = torchcrepe.filter.median(pd, 3)
+        f0 = torchcrepe.filter.mean(f0, 3)
+        f0[pd < 0.1] = 0
+        f0 = f0[0].cpu().numpy()
+
+        return f0
+
+
+class FCPE:
+    def __init__(self, device, sample_rate=16000, hop_size=160):
+        self.device = device
+        self.sample_rate = sample_rate
+        self.hop_size = hop_size
+        self.model = spawn_bundled_infer_model(self.device)
+
+    def get_f0(self, x, p_len=None, filter_radius=0.006):
+        if p_len is None:
+            p_len = x.shape[0] // self.hop_size
+
+        if not torch.is_tensor(x):
+            x = torch.from_numpy(x)
+
+        f0 = (
+            self.model.infer(
+                x.float().to(self.device).unsqueeze(0),
+                sr=self.sample_rate,
+                decoder_mode="local_argmax",
+                threshold=filter_radius,
+            )
+            .squeeze()
+            .cpu()
+            .numpy()
+        )
+
+        return f0