Update infer/lib/predictors/Generator.py

infer/lib/predictors/Generator.py

@@ -13,10 +13,10 @@ from librosa import yin, pyin, piptrack
 
 sys.path.append(os.getcwd())
 
-from main.library.predictors.CREPE.filter import mean, median
-from main.library.predictors.WORLD.SWIPE import swipe, stonemask
-from main.app.variables import config, configs, logger, translations
-from main.library.utils import autotune_f0, proposal_f0_up_key, circular_write
+from infer.lib.predictors.CREPE.filter import mean, median
+from infer.lib.predictors.WORLD.SWIPE import swipe, stonemask
+from infer.lib.variables import config, configs, logger, translations
+from infer.lib.utils import autotune_f0, proposal_f0_up_key, circular_write
 
 @nb.jit(nopython=True)
 def post_process(
@@ -338,26 +338,7 @@ class Generator:
                 mode=f0_method,
                 filter_radius=filter_radius
             )
-        elif "swipe" in f0_method:
-            f0 = self.get_f0_swipe(
-                x, 
-                p_len, 
-                filter_radius=filter_radius, 
-                use_stonemask="stonemask" in f0_method
-            )
-        elif "penn" in f0_method:
-            f0 = (
-                self.get_f0_mangio_penn(
-                    x, 
-                    p_len
-                )
-            ) if f0_method.split("-")[0] == "mangio" else (
-                self.get_f0_penn(
-                    x, 
-                    p_len, 
-                    filter_radius=filter_radius
-                )
-            )
+        
         elif "djcm" in f0_method:
             f0 = self.get_f0_djcm(
                 x, 
@@ -366,17 +347,7 @@ class Generator:
                 svs="svs" in f0_method, 
                 filter_radius=filter_radius
             )
-        elif "pesto" in f0_method:
-            f0 = self.get_f0_pesto(
-                x, 
-                p_len
-            )
-        elif "swift" in f0_method:
-            f0 = self.get_f0_swift(
-                x, 
-                p_len, 
-                filter_radius=filter_radius
-            )
+        
         else:
             raise ValueError(translations["option_not_valid"])
         
@@ -470,7 +441,7 @@ class Generator:
     
     def get_f0_mangio_crepe(self, x, p_len, model="full"):
         if not hasattr(self, "mangio_crepe"):
-            from main.library.predictors.CREPE.CREPE import CREPE
+            from infer.lib.predictors.CREPE.CREPE import CREPE
 
             self.mangio_crepe = CREPE(
                 os.path.join(
@@ -502,7 +473,7 @@ class Generator:
     
     def get_f0_crepe(self, x, p_len, model="full", filter_radius=3):
         if not hasattr(self, "crepe"):
-            from main.library.predictors.CREPE.CREPE import CREPE
+            from infer.lib.predictors.CREPE.CREPE import CREPE
 
             self.crepe = CREPE(
                 os.path.join(
@@ -531,7 +502,7 @@ class Generator:
     
     def get_f0_fcpe(self, x, p_len, legacy=False, previous=False, filter_radius=3):
         if not hasattr(self, "fcpe"): 
-            from main.library.predictors.FCPE.FCPE import FCPE
+            from infer.lib.predictors.FCPE.FCPE import FCPE
 
             self.fcpe = FCPE(
                 configs, 
@@ -566,7 +537,7 @@ class Generator:
     
     def get_f0_rmvpe(self, x, p_len, clipping=False, filter_radius=3, hpa=False, previous=False):
         if not hasattr(self, "rmvpe"): 
-            from main.library.predictors.RMVPE.RMVPE import RMVPE
+            from infer.lib.predictors.RMVPE.RMVPE import RMVPE
 
             self.rmvpe = RMVPE(
                 os.path.join(
@@ -605,60 +576,6 @@ class Generator:
         if self.predictor_onnx and self.delete_predictor_onnx: del self.rmvpe.model, self.rmvpe
         return self._resize_f0(f0, p_len)
     
-    def get_f0_pyworld(self, x, p_len, filter_radius, model="harvest", use_stonemask=True):
-        if not hasattr(self, "pw"): 
-            from main.library.predictors.WORLD.WORLD import PYWORLD
-
-            self.pw = PYWORLD(
-                os.path.join(configs["predictors_path"], "world"), 
-                os.path.join(configs["binary_path"], "world.bin")
-            )
-
-        x = x.astype(np.double)
-        pw_fn = self.pw.harvest if model == "harvest" else self.pw.dio
-
-        f0, t = pw_fn(
-            x, 
-            fs=self.sample_rate, 
-            f0_ceil=self.f0_max, 
-            f0_floor=self.f0_min, 
-            frame_period=1000 * self.window / self.sample_rate
-        )
-
-        if use_stonemask:
-            f0 = self.pw.stonemask(
-                x, 
-                self.sample_rate, 
-                t, 
-                f0
-            )
-
-        if filter_radius > 2 and model == "harvest": f0 = medfilt(f0, filter_radius)
-        elif model == "dio":
-            for index, pitch in enumerate(f0):
-                f0[index] = round(pitch, 1)
-
-        return self._resize_f0(f0, p_len)
-    
-    def get_f0_swipe(self, x, p_len, filter_radius=3, use_stonemask=True):
-        f0, t = swipe(
-            x.astype(np.float32), 
-            self.sample_rate, 
-            f0_floor=self.f0_min, 
-            f0_ceil=self.f0_max, 
-            frame_period=1000 * self.window / self.sample_rate,
-            sTHR=filter_radius / 10
-        )
-
-        if use_stonemask:
-            f0 = stonemask(
-                x, 
-                self.sample_rate, 
-                t, 
-                f0
-            )
-
-        return self._resize_f0(f0, p_len)
     
     def get_f0_librosa(self, x, p_len, mode="yin", filter_radius=3):
         if mode != "piptrack":
@@ -689,70 +606,8 @@ class Generator:
 
         return self._resize_f0(f0, p_len)
 
-    def get_f0_penn(self, x, p_len, filter_radius=3):
-        if not hasattr(self, "penn"):
-            from main.library.predictors.PENN.PENN import PENN
-
-            self.penn = PENN(
-                os.path.join(
-                    configs["predictors_path"], 
-                    f"fcn.{'onnx' if self.predictor_onnx else 'pt'}"
-                ), 
-                hop_length=self.window // 2, 
-                batch_size=self.batch_size // 2, 
-                f0_min=self.f0_min, 
-                f0_max=self.f0_max, 
-                sample_rate=self.sample_rate, 
-                device=self.device, 
-                providers=self.providers, 
-                onnx=self.predictor_onnx, 
-            )
-
-        f0, pd = self.penn.compute_f0(torch.tensor(np.copy((x)))[None].float())
-
-        if self.predictor_onnx and self.delete_predictor_onnx: 
-            del self.penn.model, self.penn.decoder
-            del self.penn.resample_audio, self.penn
-
-        f0, pd = mean(f0, filter_radius), median(pd, filter_radius)
-        f0[pd < 0.1] = 0
-
-        return self._resize_f0(f0[0].cpu().numpy(), p_len)
-
-    def get_f0_mangio_penn(self, x, p_len):
-        if not hasattr(self, "mangio_penn"):
-            from main.library.predictors.PENN.PENN import PENN
-
-            self.mangio_penn = PENN(
-                os.path.join(
-                    configs["predictors_path"], 
-                    f"fcn.{'onnx' if self.predictor_onnx else 'pt'}"
-                ), 
-                hop_length=self.hop_length // 2, 
-                batch_size=self.hop_length, 
-                f0_min=self.f0_min, 
-                f0_max=self.f0_max, 
-                sample_rate=self.sample_rate, 
-                device=self.device, 
-                providers=self.providers, 
-                onnx=self.predictor_onnx, 
-                interp_unvoiced_at=0.1
-            )
-
-        x = x.astype(np.float32)
-        x /= np.quantile(np.abs(x), 0.999)
-
-        audio = torch.from_numpy(x).to(self.device, copy=True).unsqueeze(dim=0)
-        if audio.ndim == 2 and audio.shape[0] > 1: audio = audio.mean(dim=0, keepdim=True).detach()
-
-        f0 = self.mangio_penn.compute_f0(audio.detach())
-
-        if self.predictor_onnx and self.delete_predictor_onnx: 
-            del self.mangio_penn.model, self.mangio_penn.decoder
-            del self.mangio_penn.resample_audio, self.mangio_penn
-
-        return self._resize_f0(f0.squeeze(0).cpu().float().numpy(), p_len)
-
+    
+    
     def get_f0_djcm(self, x, p_len, clipping=False, svs=False, filter_radius=3):
         if not hasattr(self, "djcm"): 
             from main.library.predictors.DJCM.DJCM import DJCM
@@ -792,48 +647,5 @@ class Generator:
         if self.predictor_onnx and self.delete_predictor_onnx: del self.djcm.model, self.djcm
         return self._resize_f0(f0, p_len)
     
-    def get_f0_swift(self, x, p_len, filter_radius=3):
-        if not hasattr(self, "swift"): 
-            from main.library.predictors.SWIFT.SWIFT import SWIFT
-
-            self.swift = SWIFT(
-                os.path.join(
-                    configs["predictors_path"], 
-                    "swift.onnx"
-                ), 
-                fmin=self.f0_min, 
-                fmax=self.f0_max, 
-                confidence_threshold=filter_radius / 4 + 0.137
-            )
-
-        pitch_hz, _, _ = self.swift.detect_from_array(x, self.sample_rate)
-        return self._resize_f0(pitch_hz, p_len)
-
-    def get_f0_pesto(self, x, p_len):
-        if not hasattr(self, "pesto"):
-            from main.library.predictors.PESTO.PESTO import PESTO
-
-            self.pesto = PESTO(
-                os.path.join(
-                    configs["predictors_path"], 
-                    f"pesto.{'onnx' if self.predictor_onnx else 'pt'}"
-                ), 
-                step_size=1000 * self.window / self.sample_rate, 
-                reduction = "alwa", 
-                num_chunks=1, 
-                sample_rate=self.sample_rate, 
-                device=self.device, 
-                providers=self.providers, 
-                onnx=self.predictor_onnx
-            )
-
-        x = x.astype(np.float32)
-        x /= np.quantile(np.abs(x), 0.999)
-
-        audio = torch.from_numpy(x).to(self.device, copy=True).unsqueeze(dim=0)
-        if audio.ndim == 2 and audio.shape[0] > 1: audio = audio.mean(dim=0, keepdim=True).detach()
-
-        f0 = self.pesto.compute_f0(audio.detach())[0]
-        if self.predictor_onnx and self.delete_predictor_onnx: del self.pesto.model, self.pesto
-
-        return self._resize_f0(f0.squeeze(0).cpu().float().numpy(), p_len)
+    
+