WIP:common sample

server/const.py

@@ -107,9 +107,9 @@ def getSampleJsonAndModelIds(mode: RVCSampleMode):
             "https://huggingface.co/wok000/vcclient_model/raw/main/samples_0003_t2.json",
             "https://huggingface.co/wok000/vcclient_model/raw/main/samples_0003_o2.json",
         ], [
-            ("TokinaShigure_o", {"useIndex": True}),
+            # ("TokinaShigure_o", {"useIndex": True}),
-            ("KikotoMahiro_o", {"useIndex": False}),
+            # ("KikotoMahiro_o", {"useIndex": False}),
-            ("Amitaro_o", {"useIndex": False}),
+            # ("Amitaro_o", {"useIndex": False}),
             ("Tsukuyomi-chan_o", {"useIndex": False}),
         ]
     elif mode == "testOfficial":

server/voice_changer/RVC/ModelSlot.py

@@ -1,28 +0,0 @@
-from const import EnumInferenceTypes, EnumEmbedderTypes
-
-from dataclasses import dataclass
-
-
-@dataclass
-class ModelSlot:
-    modelFile: str = ""
-    indexFile: str = ""
-    defaultTune: int = 0
-    defaultIndexRatio: int = 1
-    defaultProtect: float = 0.5
-    isONNX: bool = False
-    modelType: str = EnumInferenceTypes.pyTorchRVC.value
-    samplingRate: int = -1
-    f0: bool = True
-    embChannels: int = 256
-    embOutputLayer: int = 9
-    useFinalProj: bool = True
-    deprecated: bool = False
-    embedder: str = EnumEmbedderTypes.hubert.value
-
-    name: str = ""
-    description: str = ""
-    credit: str = ""
-    termsOfUseUrl: str = ""
-    sampleId: str = ""
-    iconFile: str = ""

server/voice_changer/RVC/ModelSlotGenerator.py

@@ -1,10 +1,11 @@
 from const import EnumEmbedderTypes, EnumInferenceTypes
-from voice_changer.RVC.ModelSlot import ModelSlot
 
 import torch
 import onnxruntime
 import json
 
+from data.ModelSlot import ModelSlot
+
 
 def _setInfoByPytorch(slot: ModelSlot):
     cpt = torch.load(slot.modelFile, map_location="cpu")
@@ -15,22 +16,14 @@ def _setInfoByPytorch(slot: ModelSlot):
         slot.f0 = True if cpt["f0"] == 1 else False
         version = cpt.get("version", "v1")
         if version is None or version == "v1":
-            slot.modelType = (
+            slot.modelType = EnumInferenceTypes.pyTorchRVC.value if slot.f0 else EnumInferenceTypes.pyTorchRVCNono.value
-                EnumInferenceTypes.pyTorchRVC.value
-                if slot.f0
-                else EnumInferenceTypes.pyTorchRVCNono.value
-            )
             slot.embChannels = 256
             slot.embOutputLayer = 9
             slot.useFinalProj = True
             slot.embedder = EnumEmbedderTypes.hubert.value
             print("[Voice Changer] Official Model(pyTorch) : v1")
         else:
-            slot.modelType = (
+            slot.modelType = EnumInferenceTypes.pyTorchRVCv2.value if slot.f0 else EnumInferenceTypes.pyTorchRVCv2Nono.value
-                EnumInferenceTypes.pyTorchRVCv2.value
-                if slot.f0
-                else EnumInferenceTypes.pyTorchRVCv2Nono.value
-            )
             slot.embChannels = 768
             slot.embOutputLayer = 12
             slot.useFinalProj = False
@@ -40,37 +33,21 @@ def _setInfoByPytorch(slot: ModelSlot):
     else:
         # DDPN RVC
         slot.f0 = True if cpt["f0"] == 1 else False
-        slot.modelType = (
+        slot.modelType = EnumInferenceTypes.pyTorchWebUI.value if slot.f0 else EnumInferenceTypes.pyTorchWebUINono.value
-            EnumInferenceTypes.pyTorchWebUI.value
-            if slot.f0
-            else EnumInferenceTypes.pyTorchWebUINono.value
-        )
         slot.embChannels = cpt["config"][17]
-        slot.embOutputLayer = (
+        slot.embOutputLayer = cpt["embedder_output_layer"] if "embedder_output_layer" in cpt else 9
-            cpt["embedder_output_layer"] if "embedder_output_layer" in cpt else 9
-        )
         if slot.embChannels == 256:
             slot.useFinalProj = True
         else:
             slot.useFinalProj = False
 
         # DDPNモデルの情報を表示
-        if (
+        if slot.embChannels == 256 and slot.embOutputLayer == 9 and slot.useFinalProj is True:
-            slot.embChannels == 256
-            and slot.embOutputLayer == 9
-            and slot.useFinalProj is True
-        ):
             print("[Voice Changer] DDPN Model(pyTorch) : Official v1 like")
-        elif (
+        elif slot.embChannels == 768 and slot.embOutputLayer == 12 and slot.useFinalProj is False:
-            slot.embChannels == 768
-            and slot.embOutputLayer == 12
-            and slot.useFinalProj is False
-        ):
             print("[Voice Changer] DDPN Model(pyTorch): Official v2 like")
         else:
-            print(
+            print(f"[Voice Changer] DDPN Model(pyTorch): ch:{slot.embChannels}, L:{slot.embOutputLayer}, FP:{slot.useFinalProj}")
-                f"[Voice Changer] DDPN Model(pyTorch): ch:{slot.embChannels}, L:{slot.embOutputLayer}, FP:{slot.useFinalProj}"
-            )
 
         slot.embedder = cpt["embedder_name"]
         if slot.embedder.endswith("768"):
@@ -91,9 +68,7 @@ def _setInfoByPytorch(slot: ModelSlot):
 
 
 def _setInfoByONNX(slot: ModelSlot):
-    tmp_onnx_session = onnxruntime.InferenceSession(
+    tmp_onnx_session = onnxruntime.InferenceSession(slot.modelFile, providers=["CPUExecutionProvider"])
-        slot.modelFile, providers=["CPUExecutionProvider"]
-    )
     modelmeta = tmp_onnx_session.get_modelmeta()
     try:
         metadata = json.loads(modelmeta.custom_metadata_map["metadata"])
@@ -101,16 +76,8 @@ def _setInfoByONNX(slot: ModelSlot):
         # slot.modelType = metadata["modelType"]
         slot.embChannels = metadata["embChannels"]
 
-        slot.embOutputLayer = (
+        slot.embOutputLayer = metadata["embOutputLayer"] if "embOutputLayer" in metadata else 9
-            metadata["embOutputLayer"] if "embOutputLayer" in metadata else 9
+        slot.useFinalProj = metadata["useFinalProj"] if "useFinalProj" in metadata else True if slot.embChannels == 256 else False
-        )
-        slot.useFinalProj = (
-            metadata["useFinalProj"]
-            if "useFinalProj" in metadata
-            else True
-            if slot.embChannels == 256
-            else False
-        )
 
         if slot.embChannels == 256:
             slot.useFinalProj = True
@@ -118,22 +85,12 @@ def _setInfoByONNX(slot: ModelSlot):
             slot.useFinalProj = False
 
         # ONNXモデルの情報を表示
-        if (
+        if slot.embChannels == 256 and slot.embOutputLayer == 9 and slot.useFinalProj is True:
-            slot.embChannels == 256
-            and slot.embOutputLayer == 9
-            and slot.useFinalProj is True
-        ):
             print("[Voice Changer] ONNX Model: Official v1 like")
-        elif (
+        elif slot.embChannels == 768 and slot.embOutputLayer == 12 and slot.useFinalProj is False:
-            slot.embChannels == 768
-            and slot.embOutputLayer == 12
-            and slot.useFinalProj is False
-        ):
             print("[Voice Changer] ONNX Model: Official v2 like")
         else:
-            print(
+            print(f"[Voice Changer] ONNX Model: ch:{slot.embChannels}, L:{slot.embOutputLayer}, FP:{slot.useFinalProj}")
-                f"[Voice Changer] ONNX Model: ch:{slot.embChannels}, L:{slot.embOutputLayer}, FP:{slot.useFinalProj}"
-            )
 
         if "embedder" not in metadata:
             slot.embedder = EnumEmbedderTypes.hubert.value
@@ -149,11 +106,7 @@ def _setInfoByONNX(slot: ModelSlot):
         #     raise RuntimeError("[Voice Changer][setInfoByONNX] unknown embedder")
 
         slot.f0 = metadata["f0"]
-        slot.modelType = (
+        slot.modelType = EnumInferenceTypes.onnxRVC.value if slot.f0 else EnumInferenceTypes.onnxRVCNono.value
-            EnumInferenceTypes.onnxRVC.value
-            if slot.f0
-            else EnumInferenceTypes.onnxRVCNono.value
-        )
         slot.samplingRate = metadata["samplingRate"]
         slot.deprecated = False
 

server/voice_changer/RVC/RVC.py

@@ -5,10 +5,8 @@ from typing import cast
 import numpy as np
 import torch
 import torchaudio
-from data.ModelSlot import loadAllSlotInfo
+from data.ModelSlot import RVCModelSlot, loadAllSlotInfo
 from utils.downloader.SampleDownloader import getSampleInfos
-from voice_changer.RVC.ModelSlot import ModelSlot
-from voice_changer.RVC.SampleDownloader import downloadModelFiles
 
 
 # avoiding parse arg error in RVC
@@ -102,41 +100,16 @@ class RVC:
     def loadModel(self, props: LoadModelParams):
         target_slot_idx = props.slot
         params = props.params
-        slotInfo: ModelSlot = ModelSlot()
+        slotInfo: RVCModelSlot = RVCModelSlot()
 
         print("loadModel", params)
-        # サンプルが指定されたときはダウンロードしてメタデータをでっちあげる
+        slotInfo.modelFile = params["files"]["rvcModel"]
-        if len(params["sampleId"]) > 0:
+        slotInfo.indexFile = params["files"]["rvcIndex"] if "rvcIndex" in params["files"] else None
-            sampleId = params["sampleId"]
-            sampleInfo = self.getSampleInfo(sampleId)
-            useIndex = params["rvcIndexDownload"]
-
-            if sampleInfo is None:
-                print("[Voice Changer] sampleInfo is None")
-                return
-            modelPath, indexPath, iconPath = downloadModelFiles(sampleInfo, useIndex)
-            slotInfo.modelFile = modelPath
-            if indexPath is not None:
-                slotInfo.indexFile = indexPath
-            if iconPath is not None:
-                slotInfo.iconFile = iconPath
-
-            slotInfo.sampleId = sampleInfo.id
-            slotInfo.credit = sampleInfo.credit
-            slotInfo.description = sampleInfo.description
-            slotInfo.name = sampleInfo.name
-            slotInfo.termsOfUseUrl = sampleInfo.termsOfUseUrl
-
-            # slotInfo.samplingRate = sampleInfo.sampleRate
-            # slotInfo.modelType = sampleInfo.modelType
-            # slotInfo.f0 = sampleInfo.f0
-        else:
-            slotInfo.modelFile = params["files"]["rvcModel"]
-            slotInfo.indexFile = params["files"]["rvcIndex"] if "rvcIndex" in params["files"] else None
 
         slotInfo.defaultTune = params["defaultTune"]
         slotInfo.defaultIndexRatio = params["defaultIndexRatio"]
         slotInfo.defaultProtect = params["defaultProtect"]
+        slotInfo.voiceChangerType = "RVC"
         slotInfo.isONNX = slotInfo.modelFile.endswith(".onnx")
 
         if slotInfo.isONNX:

server/voice_changer/RVC/SampleDownloader.py

@@ -1,164 +0,0 @@
-from concurrent.futures import ThreadPoolExecutor
-from dataclasses import asdict
-import os
-from const import RVC_MODEL_DIRNAME, TMP_DIR
-from Downloader import download, download_no_tqdm
-from ModelSample import RVCModelSample, getModelSamples
-import json
-
-from voice_changer.RVC.ModelSlot import ModelSlot
-from voice_changer.RVC.ModelSlotGenerator import _setInfoByONNX, _setInfoByPytorch
-
-
-def checkRvcModelExist(model_dir: str):
-    rvcModelDir = os.path.join(model_dir, RVC_MODEL_DIRNAME)
-    if not os.path.exists(rvcModelDir):
-        return False
-    return True
-
-
-def downloadInitialSampleModels(sampleJsons: list[str], sampleModelIds: list[str], model_dir: str):
-    sampleModels = getModelSamples(sampleJsons, "RVC")
-    if sampleModels is None:
-        return
-
-    downloadParams = []
-    slot_count = 0
-    line_num = 0
-    for initSampleId in sampleModelIds:
-        # 初期サンプルをサーチ
-        match = False
-        for sample in sampleModels:
-            if sample.id == initSampleId[0]:
-                match = True
-                break
-        if match is False:
-            print(f"[Voice Changer] initiail sample not found. {initSampleId[0]}")
-            continue
-
-        # 検出されたら、、、
-        slotInfo: ModelSlot = ModelSlot()
-        # sampleParams: Any = {"files": {}}
-
-        slotDir = os.path.join(model_dir, RVC_MODEL_DIRNAME, str(slot_count))
-        os.makedirs(slotDir, exist_ok=True)
-        modelFilePath = os.path.join(
-            slotDir,
-            os.path.basename(sample.modelUrl),
-        )
-        downloadParams.append(
-            {
-                "url": sample.modelUrl,
-                "saveTo": modelFilePath,
-                "position": line_num,
-            }
-        )
-        slotInfo.modelFile = modelFilePath
-        line_num += 1
-
-        if initSampleId[1] is True and hasattr(sample, "indexUrl") and sample.indexUrl != "":
-            indexPath = os.path.join(
-                slotDir,
-                os.path.basename(sample.indexUrl),
-            )
-            downloadParams.append(
-                {
-                    "url": sample.indexUrl,
-                    "saveTo": indexPath,
-                    "position": line_num,
-                }
-            )
-            slotInfo.indexFile = indexPath
-            line_num += 1
-        if hasattr(sample, "icon") and sample.icon != "":
-            iconPath = os.path.join(
-                slotDir,
-                os.path.basename(sample.icon),
-            )
-            downloadParams.append(
-                {
-                    "url": sample.icon,
-                    "saveTo": iconPath,
-                    "position": line_num,
-                }
-            )
-            slotInfo.iconFile = iconPath
-            line_num += 1
-
-        slotInfo.sampleId = sample.id
-        slotInfo.credit = sample.credit
-        slotInfo.description = sample.description
-        slotInfo.name = sample.name
-        slotInfo.termsOfUseUrl = sample.termsOfUseUrl
-        slotInfo.defaultTune = 0
-        slotInfo.defaultIndexRatio = 0.0
-        slotInfo.defaultProtect = 0.5
-        slotInfo.isONNX = slotInfo.modelFile.endswith(".onnx")
-
-        # この時点ではまだファイルはダウンロードされていない
-        # if slotInfo.isONNX:
-        #     _setInfoByONNX(slotInfo)
-        # else:
-        #     _setInfoByPytorch(slotInfo)
-
-        json.dump(asdict(slotInfo), open(os.path.join(slotDir, "params.json"), "w"))
-        slot_count += 1
-
-    # ダウンロード
-    print("[Voice Changer] Downloading model files...")
-    with ThreadPoolExecutor() as pool:
-        pool.map(download, downloadParams)
-
-    # メタデータ作成
-    print("[Voice Changer] Generating metadata...")
-    for slotId in range(slot_count):
-        slotDir = os.path.join(model_dir, RVC_MODEL_DIRNAME, str(slotId))
-        jsonDict = json.load(open(os.path.join(slotDir, "params.json")))
-        slotInfo = ModelSlot(**jsonDict)
-        if slotInfo.isONNX:
-            _setInfoByONNX(slotInfo)
-        else:
-            _setInfoByPytorch(slotInfo)
-        json.dump(asdict(slotInfo), open(os.path.join(slotDir, "params.json"), "w"))
-
-
-def downloadModelFiles(sampleInfo: RVCModelSample, useIndex: bool = True):
-    downloadParams = []
-
-    modelPath = os.path.join(TMP_DIR, os.path.basename(sampleInfo.modelUrl))
-    downloadParams.append(
-        {
-            "url": sampleInfo.modelUrl,
-            "saveTo": modelPath,
-            "position": 0,
-        }
-    )
-
-    indexPath = None
-    if useIndex is True and hasattr(sampleInfo, "indexUrl") and sampleInfo.indexUrl != "":
-        print("[Voice Changer] Download sample with index.")
-        indexPath = os.path.join(TMP_DIR, os.path.basename(sampleInfo.indexUrl))
-        downloadParams.append(
-            {
-                "url": sampleInfo.indexUrl,
-                "saveTo": indexPath,
-                "position": 1,
-            }
-        )
-
-    iconPath = None
-    if hasattr(sampleInfo, "icon") and sampleInfo.icon != "":
-        iconPath = os.path.join(TMP_DIR, os.path.basename(sampleInfo.icon))
-        downloadParams.append(
-            {
-                "url": sampleInfo.icon,
-                "saveTo": iconPath,
-                "position": 2,
-            }
-        )
-
-    print("[Voice Changer] Downloading model files...", end="")
-    with ThreadPoolExecutor() as pool:
-        pool.map(download_no_tqdm, downloadParams)
-    print("")
-    return modelPath, indexPath, iconPath

server/voice_changer/RVC/onnxExporter/export2onnx.py

@@ -4,7 +4,7 @@ import torch
 from onnxsim import simplify
 import onnx
 from const import TMP_DIR, EnumInferenceTypes
-from voice_changer.RVC.ModelSlot import ModelSlot
+from data.ModelSlot import ModelSlot
 from voice_changer.RVC.deviceManager.DeviceManager import DeviceManager
 from voice_changer.RVC.onnxExporter.SynthesizerTrnMs256NSFsid_ONNX import (
     SynthesizerTrnMs256NSFsid_ONNX,
@@ -30,9 +30,7 @@ def export2onnx(gpu: int, modelSlot: ModelSlot):
     modelFile = modelSlot.modelFile
 
     output_file = os.path.splitext(os.path.basename(modelFile))[0] + ".onnx"
-    output_file_simple = (
+    output_file_simple = os.path.splitext(os.path.basename(modelFile))[0] + "_simple.onnx"
-        os.path.splitext(os.path.basename(modelFile))[0] + "_simple.onnx"
-    )
     output_path = os.path.join(TMP_DIR, output_file)
     output_path_simple = os.path.join(TMP_DIR, output_file_simple)
     metadata = {
@@ -52,9 +50,7 @@ def export2onnx(gpu: int, modelSlot: ModelSlot):
     if gpuMomory > 0:
         _export2onnx(modelFile, output_path, output_path_simple, True, metadata)
     else:
-        print(
+        print("[Voice Changer] Warning!!! onnx export with float32. maybe size is doubled.")
-            "[Voice Changer] Warning!!! onnx export with float32. maybe size is doubled."
-        )
         _export2onnx(modelFile, output_path, output_path_simple, False, metadata)
     return output_file_simple
 

server/voice_changer/RVC/pipeline/PipelineGenerator.py

@@ -1,8 +1,8 @@
 import os
 import traceback
 import faiss
+from data.ModelSlot import RVCModelSlot
 
-from voice_changer.RVC.ModelSlot import ModelSlot
 from voice_changer.RVC.deviceManager.DeviceManager import DeviceManager
 from voice_changer.RVC.embedder.EmbedderManager import EmbedderManager
 from voice_changer.RVC.inferencer.InferencerManager import InferencerManager
@@ -10,15 +10,13 @@ from voice_changer.RVC.pipeline.Pipeline import Pipeline
 from voice_changer.RVC.pitchExtractor.PitchExtractorManager import PitchExtractorManager
 
 
-def createPipeline(modelSlot: ModelSlot, gpu: int, f0Detector: str):
+def createPipeline(modelSlot: RVCModelSlot, gpu: int, f0Detector: str):
     dev = DeviceManager.get_instance().getDevice(gpu)
     half = DeviceManager.get_instance().halfPrecisionAvailable(gpu)
 
     # Inferencer 生成
     try:
-        inferencer = InferencerManager.getInferencer(
+        inferencer = InferencerManager.getInferencer(modelSlot.modelType, modelSlot.modelFile, gpu)
-            modelSlot.modelType, modelSlot.modelFile, gpu
-        )
     except Exception as e:
         print("[Voice Changer] exception! loading inferencer", e)
         traceback.print_exc()
@@ -54,7 +52,7 @@ def createPipeline(modelSlot: ModelSlot, gpu: int, f0Detector: str):
     return pipeline
 
 
-def _loadIndex(modelSlot: ModelSlot):
+def _loadIndex(modelSlot: RVCModelSlot):
     # Indexのロード
     print("[Voice Changer] Loading index...")
     # ファイル指定がない場合はNone