optimize(onnx): move infer into rvc.onnx

.gitignore

@@ -4,25 +4,10 @@ __pycache__
 *.pyd
 .venv
 /opt
-tools/aria2c/
-tools/flag.txt
-
-# Imported from huggingface.co/lj1995/VoiceConversionWebUI
-/pretrained
-/pretrained_v2
-/uvr5_weights
-hubert_base.pt
-rmvpe.onnx
-rmvpe.pt
 
 # Generated by RVC
 /logs
-/weights
 
-# To set a Python version for the project
-.tool-versions
-
-/runtime
 /assets/weights/*
 ffmpeg.*
-ffprobe.*
+ffprobe.*

configs/config.py

@@ -211,27 +211,6 @@ class Config:
             x_max = 32
         if self.dml:
             logger.info("Use DirectML instead")
-            if (
-                os.path.exists(
-                    "runtime\Lib\site-packages\onnxruntime\capi\DirectML.dll"
-                )
-                == False
-            ):
-                try:
-                    os.rename(
-                        "runtime\Lib\site-packages\onnxruntime",
-                        "runtime\Lib\site-packages\onnxruntime-cuda",
-                    )
-                except:
-                    pass
-                try:
-                    os.rename(
-                        "runtime\Lib\site-packages\onnxruntime-dml",
-                        "runtime\Lib\site-packages\onnxruntime",
-                    )
-                except:
-                    pass
-            # if self.device != "cpu":
             import torch_directml
 
             self.device = torch_directml.device(torch_directml.default_device())
@@ -239,26 +218,6 @@ class Config:
         else:
             if self.instead:
                 logger.info(f"Use {self.instead} instead")
-            if (
-                os.path.exists(
-                    "runtime\Lib\site-packages\onnxruntime\capi\onnxruntime_providers_cuda.dll"
-                )
-                == False
-            ):
-                try:
-                    os.rename(
-                        "runtime\Lib\site-packages\onnxruntime",
-                        "runtime\Lib\site-packages\onnxruntime-dml",
-                    )
-                except:
-                    pass
-                try:
-                    os.rename(
-                        "runtime\Lib\site-packages\onnxruntime-cuda",
-                        "runtime\Lib\site-packages\onnxruntime",
-                    )
-                except:
-                    pass
         logger.info(
             "Half-precision floating-point: %s, device: %s"
             % (self.is_half, self.device)

i18n/scan_i18n.py

@@ -36,15 +36,6 @@ for filename in glob.iglob("**/*.py", recursive=True):
             print(filename, len(i18n_strings))
             strings.extend(i18n_strings)
 code_keys = set(strings)
-"""
-n_i18n.py
-gui_v1.py 26
-app.py 16
-infer-web.py 147
-scan_i18n.py 0
-i18n.py 0
-lib/train/process_ckpt.py 1
-"""
 print()
 print("Total unique:", len(code_keys))
 

rvc/onnx/f0predictor/__init__.py

@@ -0,0 +1,3 @@
+from .dio import DioF0Predictor
+from .harvest import HarvestF0Predictor
+from .pm import PMF0Predictor

rvc/onnx/f0predictor/dio.py

@@ -1,7 +1,7 @@
 import numpy as np
 import pyworld
 
-from infer.lib.infer_pack.modules.F0Predictor.F0Predictor import F0Predictor
+from .f0 import F0Predictor
 
 
 class DioF0Predictor(F0Predictor):

rvc/onnx/f0predictor/harvest.py

@@ -1,7 +1,7 @@
 import numpy as np
 import pyworld
 
-from infer.lib.infer_pack.modules.F0Predictor.F0Predictor import F0Predictor
+from .f0 import F0Predictor
 
 
 class HarvestF0Predictor(F0Predictor):

rvc/onnx/f0predictor/pm.py

@@ -1,7 +1,7 @@
 import numpy as np
 import parselmouth
 
-from infer.lib.infer_pack.modules.F0Predictor.F0Predictor import F0Predictor
+from .f0 import F0Predictor
 
 
 class PMF0Predictor(F0Predictor):

rvc/onnx/infer.py

@@ -1,16 +1,13 @@
 import librosa
 import numpy as np
 import onnxruntime
-import soundfile
-
-import logging
-
-logger = logging.getLogger(__name__)
 
+from onnx.f0predictor import PMF0Predictor
+from onnx.f0predictor import HarvestF0Predictor
+from onnx.f0predictor import DioF0Predictor
 
 class ContentVec:
-    def __init__(self, vec_path="pretrained/vec-768-layer-12.onnx", device=None):
+    def __init__(self, vec_path: str, device=None):
-        logger.info("Load model(s) from {}".format(vec_path))
         if device == "cpu" or device is None:
             providers = ["CPUExecutionProvider"]
         elif device == "cuda":
@@ -25,52 +22,33 @@ class ContentVec:
         return self.forward(wav)
 
     def forward(self, wav):
-        feats = wav
+        if wav.ndim == 2:  # double channels
-        if feats.ndim == 2:  # double channels
+            wav = wav.mean(-1)
-            feats = feats.mean(-1)
+        assert wav.ndim == 1, wav.ndim
-        assert feats.ndim == 1, feats.ndim
+        wav = np.expand_dims(np.expand_dims(wav, 0), 0)
-        feats = np.expand_dims(np.expand_dims(feats, 0), 0)
+        onnx_input = {self.model.get_inputs()[0].name: wav}
-        onnx_input = {self.model.get_inputs()[0].name: feats}
         logits = self.model.run(None, onnx_input)[0]
         return logits.transpose(0, 2, 1)
 
+predicters = {
+    "pm": PMF0Predictor,
+    "harvest": HarvestF0Predictor,
+    "dio": DioF0Predictor,
+}
 
-def get_f0_predictor(f0_predictor, hop_length, sampling_rate, **kargs):
+def get_f0_predictor(f0_method, hop_length, sampling_rate):
-    if f0_predictor == "pm":
+    return predicters[f0_method](hop_length=hop_length, sampling_rate=sampling_rate)
-        from lib.infer_pack.modules.F0Predictor.PMF0Predictor import PMF0Predictor
-
-        f0_predictor_object = PMF0Predictor(
-            hop_length=hop_length, sampling_rate=sampling_rate
-        )
-    elif f0_predictor == "harvest":
-        from lib.infer_pack.modules.F0Predictor.HarvestF0Predictor import (
-            HarvestF0Predictor,
-        )
-
-        f0_predictor_object = HarvestF0Predictor(
-            hop_length=hop_length, sampling_rate=sampling_rate
-        )
-    elif f0_predictor == "dio":
-        from lib.infer_pack.modules.F0Predictor.DioF0Predictor import DioF0Predictor
-
-        f0_predictor_object = DioF0Predictor(
-            hop_length=hop_length, sampling_rate=sampling_rate
-        )
-    else:
-        raise Exception("Unknown f0 predictor")
-    return f0_predictor_object
 
 
-class OnnxRVC:
+class RVC:
     def __init__(
         self,
         model_path,
         sr=40000,
         hop_size=512,
-        vec_path="vec-768-layer-12",
+        vec_path="vec-768-layer-12.onnx",
         device="cpu",
     ):
-        vec_path = f"pretrained/{vec_path}.onnx"
         self.vec_model = ContentVec(vec_path, device)
         if device == "cpu" or device is None:
             providers = ["CPUExecutionProvider"]
@@ -97,12 +75,11 @@ class OnnxRVC:
 
     def inference(
         self,
-        raw_path,
+        wav,
+        sr,
         sid,
         f0_method="dio",
         f0_up_key=0,
-        pad_time=0.5,
-        cr_threshold=0.02,
     ):
         f0_min = 50
         f0_max = 1100
@@ -110,16 +87,14 @@ class OnnxRVC:
         f0_mel_max = 1127 * np.log(1 + f0_max / 700)
         f0_predictor = get_f0_predictor(
             f0_method,
-            hop_length=self.hop_size,
+            self.hop_size,
-            sampling_rate=self.sampling_rate,
+            self.sampling_rate,
-            threshold=cr_threshold,
         )
-        wav, sr = librosa.load(raw_path, sr=self.sampling_rate)
         org_length = len(wav)
         if org_length / sr > 50.0:
             raise RuntimeError("Reached Max Length")
 
-        wav16k = librosa.resample(wav, orig_sr=self.sampling_rate, target_sr=16000)
+        wav16k = librosa.resample(wav, orig_sr=sr, target_sr=16000)
         wav16k = wav16k
 
         hubert = self.vec_model(wav16k)

tools/onnx/onnx_inference_demo.py

@@ -1,23 +1,24 @@
 import soundfile
+import librosa
 
-from infer.lib.infer_pack.onnx_inference import OnnxRVC
+from rvc.onnx.infer import RVC
 
 hop_size = 512
 sampling_rate = 40000  # 采样率
 f0_up_key = 0  # 升降调
 sid = 0  # 角色ID
 f0_method = "dio"  # F0提取算法
-model_path = "ShirohaRVC.onnx"  # 模型的完整路径
+model_path = "exported_model.onnx"  # 模型的完整路径
-vec_name = (
+vec_path = "vec-256-layer-9.onnx" # 需要onnx的vec模型
-    "vec-256-layer-9"  # 内部自动补齐为 f"pretrained/{vec_name}.onnx" 需要onnx的vec模型
-)
 wav_path = "123.wav"  # 输入路径或ByteIO实例
 out_path = "out.wav"  # 输出路径或ByteIO实例
 
-model = OnnxRVC(
+model = RVC(
-    model_path, vec_path=vec_name, sr=sampling_rate, hop_size=hop_size, device="cuda"
+    model_path, vec_path=vec_path, sr=sampling_rate, hop_size=hop_size, device="cuda"
 )
 
-audio = model.inference(wav_path, sid, f0_method=f0_method, f0_up_key=f0_up_key)
+wav, sr = librosa.load(wav_path, sr=sampling_rate)
+
+audio = model.inference(wav, sr, sid, f0_method=f0_method, f0_up_key=f0_up_key)
 
 soundfile.write(out_path, audio, sampling_rate)

web.py

@@ -44,8 +44,6 @@ logger = logging.getLogger(__name__)
 
 tmp = os.path.join(now_dir, "TEMP")
 shutil.rmtree(tmp, ignore_errors=True)
-shutil.rmtree("%s/runtime/Lib/site-packages/infer_pack" % (now_dir), ignore_errors=True)
-shutil.rmtree("%s/runtime/Lib/site-packages/uvr5_pack" % (now_dir), ignore_errors=True)
 os.makedirs(tmp, exist_ok=True)
 os.makedirs(os.path.join(now_dir, "logs"), exist_ok=True)
 os.makedirs(os.path.join(now_dir, "assets/weights"), exist_ok=True)