wip:support m1 mac 1

client/demo/src/components/demo/904-2_SampleDownloader.tsx

@@ -1,6 +1,6 @@
 import React, { useMemo, useState } from "react";
 import { useAppState } from "../../001_provider/001_AppStateProvider";
-import { InitialFileUploadSetting } from "@dannadori/voice-changer-client-js";
+import { ModelUploadSetting } from "@dannadori/voice-changer-client-js";
 import { useMessageBuilder } from "../../hooks/useMessageBuilder";
 import { ModelSlotManagerDialogScreen } from "./904_ModelSlotManagerDialog";
 
@@ -51,15 +51,18 @@ export const SampleDownloaderScreen = (props: SampleDownloaderScreenProps) => {
         )
 
         const onDownloadSampleClicked = async (id: string) => {
-            serverSetting.fileUploadSettings[props.targetIndex] = {
+            const uploadParams: ModelUploadSetting = {
-                ...InitialFileUploadSetting,
+                voiceChangerType: "RVC",
-                rvcModel: null,
+                slot: props.targetIndex,
-                rvcIndex: null,
+                isSampleMode: true,
                 sampleId: id,
-                isSampleMode: true
+                files: [],
+                params: {
+                    rvcIndexDownload: true
+                }
             }
             try {
-                await serverSetting.loadModel(props.targetIndex)
+                await serverSetting.uploadModel(uploadParams)
             } catch (e) {
                 alert(e)
             }

client/demo/src/components/demo/904-3_FileUploader.tsx

@@ -41,6 +41,7 @@ export const FileUploaderScreen = (props: FileUploaderScreenProps) => {
             isSampleMode: false,
             sampleId: null,
             files: [],
+            params: {}
         })
     }, [props.targetIndex, voiceChangerType])
 

client/lib/src/hooks/useServerSetting.ts

@@ -51,6 +51,7 @@ export type ModelUploadSetting = {
     sampleId: string | null
 
     files: ModelFile[]
+    params: any
 }
 export type ModelFileForServer = Omit<ModelFile, "file"> & {
     name: string,
@@ -341,7 +342,6 @@ export const useServerSetting = (props: UseServerSettingProps): ServerSettingSta
         }
     }, [props.voiceChangerClient])
 
-
     // 古いアップローダ（新GUIへ以降まで、当分残しておく。）
     const loadModel = useMemo(() => {
         return async (slot: number) => {

server/downloader/SampleDownloader.py

@@ -7,7 +7,7 @@ from const import RVCSampleMode, getSampleJsonAndModelIds
 from data.ModelSample import ModelSamples, generateModelSample
 from data.ModelSlot import RVCModelSlot
 from voice_changer.ModelSlotManager import ModelSlotManager
-from voice_changer.RVC.ModelSlotGenerator import _setInfoByONNX, _setInfoByPytorch
+from voice_changer.RVC.RVCModelSlotGenerator import RVCModelSlotGenerator
 from downloader.Downloader import download, download_no_tqdm
 
 
@@ -166,7 +166,7 @@ def _downloadSamples(samples: list[ModelSamples], sampleModelIds: list[Tuple[str
         slotInfo = modelSlotManager.get_slot_info(targetSlotIndex)
         if slotInfo.voiceChangerType == "RVC":
             if slotInfo.isONNX:
-                _setInfoByONNX(slotInfo)
+                RVCModelSlotGenerator._setInfoByONNX(slotInfo)
             else:
-                _setInfoByPytorch(slotInfo)
+                RVCModelSlotGenerator._setInfoByPytorch(slotInfo)
             modelSlotManager.save_model_slot(targetSlotIndex, slotInfo)

server/restapi/MMVC_Rest_Fileuploader.py

@@ -1,7 +1,6 @@
 import sys
 import json
 import os
-import shutil
 from typing import Union
 from fastapi import APIRouter
 from fastapi.encoders import jsonable_encoder
@@ -12,7 +11,7 @@ from restapi.mods.FileUploader import upload_file, concat_file_chunks
 from voice_changer.VoiceChangerManager import VoiceChangerManager
 
 from const import MODEL_DIR, UPLOAD_DIR, ModelType
-from voice_changer.utils.LoadModelParams import LoadModelParams
+from voice_changer.utils.LoadModelParams import LoadModelParamFile, LoadModelParams
 
 
 os.makedirs(UPLOAD_DIR, exist_ok=True)
@@ -88,26 +87,11 @@ class MMVC_Rest_Fileuploader:
         try:
             paramDict = json.loads(params)
             print("paramDict", paramDict)
-            if paramDict["voiceChangerType"]:
+            loadModelparams = LoadModelParams(**paramDict)
-                # 新しいアップローダ用
+            loadModelparams.files = [LoadModelParamFile(**x) for x in paramDict["files"]]
-                print("NEW UPLOADER")
+            # print("paramDict", loadModelparams)
-                props: LoadModelParams = LoadModelParams(slot=slot, isHalf=isHalf, params=paramDict)
-            else:
-                # 古いアップローダ用
-                # Change Filepath
-                newFilesDict = {}
-                for key, val in paramDict["files"].items():
-                    if val != "-" and val != "":
-                        uploadPath = os.path.join(UPLOAD_DIR, val)
-                        storePath = os.path.join(UPLOAD_DIR, f"{slot}", val)
-                        storeDir = os.path.dirname(storePath)
-                        os.makedirs(storeDir, exist_ok=True)
-                        shutil.move(uploadPath, storePath)
-                        newFilesDict[key] = storePath
-                paramDict["files"] = newFilesDict
-                props = LoadModelParams(slot=slot, isHalf=isHalf, params=paramDict)
 
-            info = self.voiceChangerManager.loadModel(props)
+            info = self.voiceChangerManager.loadModel(loadModelparams)
             json_compatible_item_data = jsonable_encoder(info)
             return JSONResponse(content=json_compatible_item_data)
         except Exception as e:

server/restapi/mods/Trainer_Model.py

@@ -1,14 +0,0 @@
-
-from fastapi.responses import FileResponse
-import os
-
-
-def mod_get_model(modelFile: str):
-    modelPath = os.path.join("MMVC_Trainer/logs", modelFile)
-    return FileResponse(path=modelPath)
-
-
-def mod_delete_model(modelFile: str):
-    modelPath = os.path.join("MMVC_Trainer/logs", modelFile)
-    os.remove(modelPath)
-    return {"Model deleted": f"{modelFile}"}

server/restapi/mods/Trainer_Models.py

@@ -1,23 +0,0 @@
-
-from fastapi.responses import JSONResponse
-from fastapi.encoders import jsonable_encoder
-from restapi.utils.files import get_file_list
-import os
-
-def mod_get_models():
-    gModels = get_file_list(f'MMVC_Trainer/logs/G*.pth')
-    dModels = get_file_list(f'MMVC_Trainer/logs/D*.pth')
-    configs = get_file_list(f'MMVC_Trainer/logs/config.json')
-    models = []
-    models.extend(gModels)
-    models.extend(dModels)
-    models.extend(configs)
-    models = [ os.path.basename(x) for x in models]
-
-    models = sorted(models)
-    data = {
-            "models":models
-            }
-    json_compatible_item_data = jsonable_encoder(data)
-    return JSONResponse(content=json_compatible_item_data)
-

server/restapi/mods/Trainer_MultiSpeakerSetting.py

@@ -1,26 +0,0 @@
-from fastapi.responses import JSONResponse
-from fastapi.encoders import jsonable_encoder
-import os
-
-MULTI_SPEAKER_SETTING_PATH = "MMVC_Trainer/dataset/multi_speaker_correspondence.txt"
-def mod_get_multi_speaker_setting():
-    data = {}
-    if os.path.isfile(MULTI_SPEAKER_SETTING_PATH) == False:
-        with open(MULTI_SPEAKER_SETTING_PATH, "w") as f:
-            f.write("")
-            f.flush()
-        f.close()
-
-    with open(MULTI_SPEAKER_SETTING_PATH, "r") as f:
-        setting = f.read()
-        data["multi_speaker_setting"] = setting
-    json_compatible_item_data = jsonable_encoder(data)
-    return JSONResponse(content=json_compatible_item_data)
-
-
-def mod_post_multi_speaker_setting(setting:str):
-    with open(MULTI_SPEAKER_SETTING_PATH, "w") as f:
-        f.write(setting)
-        f.flush()
-    f.close()
-    return {"Write Multispeaker setting": f"{setting}"}

server/restapi/mods/Trainer_Speaker.py

@@ -1,15 +0,0 @@
-import shutil
-from restapi.mods.Trainer_MultiSpeakerSetting import MULTI_SPEAKER_SETTING_PATH
-
-def mod_delete_speaker(speaker:str):
-        shutil.rmtree(f"MMVC_Trainer/dataset/textful/{speaker}")
-
-        with open(MULTI_SPEAKER_SETTING_PATH, "r") as f:
-            setting = f.readlines()
-
-        filtered = filter(lambda x: x.startswith(f"{speaker}|")==False, setting)
-        with open(MULTI_SPEAKER_SETTING_PATH, "w") as f:
-            f.writelines(list(filtered))
-            f.flush()
-        f.close()
-        return {"Speaker deleted": f"{speaker}"}

server/restapi/mods/Trainer_Speaker_Voice.py

@@ -1,28 +0,0 @@
-from fastapi.responses import JSONResponse
-from fastapi.encoders import jsonable_encoder
-import os,  base64
-
-def mod_get_speaker_voice(speaker:str, voice:str):
-        wav_file = f'MMVC_Trainer/dataset/textful/{speaker}/wav/{voice}.wav'
-        text_file = f'MMVC_Trainer/dataset/textful/{speaker}/text/{voice}.txt'
-        readable_text_file = f'MMVC_Trainer/dataset/textful/{speaker}/readable_text/{voice}.txt'
-
-        data = {}
-        if os.path.exists(wav_file):
-            with open(wav_file, "rb") as f:
-                wav_data = f.read()
-            wav_data_base64 = base64.b64encode(wav_data).decode('utf-8')
-            data["wav"] = wav_data_base64
-
-
-        if os.path.exists(text_file):
-            with open(text_file, "r") as f:
-                text_data = f.read()
-            data["text"] = text_data
-
-        if os.path.exists(readable_text_file):
-            with open(readable_text_file, "r") as f:
-                text_data = f.read()
-            data["readable_text"] = text_data
-        json_compatible_item_data = jsonable_encoder(data)
-        return JSONResponse(content=json_compatible_item_data)

server/restapi/mods/Trainer_Speaker_Voices.py

@@ -1,22 +0,0 @@
-from fastapi.responses import JSONResponse
-from fastapi.encoders import jsonable_encoder
-from restapi.utils.files import get_file_list
-import os
-
-def mod_get_speaker_voices(speaker:str):
-        voices = get_file_list(f'MMVC_Trainer/dataset/textful/{speaker}/wav/*.wav')
-
-        texts = get_file_list(f'MMVC_Trainer/dataset/textful/{speaker}/text/*.txt')
-
-        readable_texts = get_file_list(f'MMVC_Trainer/dataset/textful/{speaker}/readable_text/*.txt')
-
-        items = voices
-        items.extend(texts)
-        items.extend(readable_texts)
-        items = [ os.path.splitext(os.path.basename(x))[0] for x in items]
-        items = sorted(set(items))
-        data = {
-                "voices":items
-                }
-        json_compatible_item_data = jsonable_encoder(data)
-        return JSONResponse(content=json_compatible_item_data)

server/restapi/mods/Trainer_Speakers.py

@@ -1,15 +0,0 @@
-from fastapi.responses import JSONResponse
-from fastapi.encoders import jsonable_encoder
-from restapi.utils.files import get_dir_list
-import os
-# CreateはFileUploaderで実装。
-
-def mod_get_speakers():
-    os.makedirs("MMVC_Trainer/dataset/textful", exist_ok=True)
-    speakers = get_dir_list("MMVC_Trainer/dataset/textful/")
-    
-    data = {
-        "speakers":sorted(speakers)
-    }
-    json_compatible_item_data = jsonable_encoder(data)
-    return JSONResponse(content=json_compatible_item_data)

server/restapi/mods/Trainer_Training.py

@@ -1,176 +0,0 @@
-import subprocess,os
-from restapi.utils.files import get_file_list
-from fastapi.responses import JSONResponse
-from fastapi.encoders import jsonable_encoder
-
-LOG_DIR = "info"
-train_proc = None
-
-SUCCESS = 0
-ERROR = -1
-### Submodule for Pre train
-def sync_exec(cmd:str, log_path:str, cwd=None):
-    shortCmdStr = cmd[:20]
-    try:
-        with open(log_path, 'w') as log_file:
-            if cwd == None:
-                proc = subprocess.run(cmd, shell=True, text=True, stdout=log_file, stderr=log_file)
-            else:
-                proc = subprocess.run(cmd, shell=True, text=True, stdout=log_file, stderr=log_file, cwd=cwd)
-            print(f"{shortCmdStr} returncode:{proc.returncode}")
-            if proc.returncode != 0:
-                print(f"{shortCmdStr} exception:")
-                return (ERROR, f"returncode:{proc.returncode}")
-    except Exception as e:
-        print(f"{shortCmdStr} exception:", str(e))
-        return (ERROR, str(e))
-    return (SUCCESS, "success")
-
-def sync_exec_with_stdout(cmd:str, log_path:str):
-    shortCmdStr = cmd[:20]
-    try:
-        with open(log_path, 'w') as log_file:
-            proc = subprocess.run(cmd, shell=True, text=True, stdout=subprocess.PIPE,
-            stderr=log_file, cwd="MMVC_Trainer")
-            print(f"STDOUT{shortCmdStr}",proc.stdout)
-    except Exception as e:
-        print(f"{shortCmdStr} exception:", str(e))
-        return (ERROR, str(e))
-    return (SUCCESS, proc.stdout)
-
-
-def create_dataset():
-    cmd = "python3 create_dataset_jtalk.py -f train_config -s 24000 -m dataset/multi_speaker_correspondence.txt"
-    log_file = os.path.join(LOG_DIR, "log_create_dataset_jtalk.txt")
-    res = sync_exec(cmd, log_file, "MMVC_Trainer")
-    return res
-
-def set_batch_size(batch:int):
-    cmd = "sed -i 's/\"batch_size\": [0-9]*/\"batch_size\": " + str(batch) + "/' MMVC_Trainer/configs/baseconfig.json"
-    log_file = os.path.join(LOG_DIR, "log_set_batch_size.txt")
-    res = sync_exec(cmd, log_file)
-    return res
-
-def set_dummy_device_count():
-    cmd = 'sed -ie "s/torch.cuda.device_count()/1/" MMVC_Trainer/train_ms.py'
-    log_file = os.path.join(LOG_DIR, "log_set_dummy_device_count.txt")
-    res = sync_exec(cmd, log_file)
-    return res
-
-### Submodule for Train 
-def exec_training(enable_finetuning:bool, GModel:str, DModel:str):
-    global train_proc
-    log_file = os.path.join(LOG_DIR, "training.txt")
-
-    # トレーニング開始確認(二重起動回避)
-    if train_proc != None:
-        status = train_proc.poll()
-        if status != None:
-            print("Training have ended.", status)
-            train_proc = None
-        else:
-            print("Training have stated.")
-            return (ERROR, "Training have started")
-
-    try:
-        with open(log_file, 'w') as log_file:
-            if enable_finetuning == True:
-                GModelPath = os.path.join("logs", GModel) # 実行時にcwdを指定しているのでフォルダはlogsでよい。
-                DModelPath = os.path.join("logs", DModel)
-                cmd = f'python3 train_ms.py -c configs/train_config.json -m ./ -fg {GModelPath} -fd {DModelPath}'
-            else:
-                cmd = 'python3 train_ms.py -c configs/train_config.json -m ./'
-            print("exec:",cmd)
-            train_proc = subprocess.Popen("exec "+cmd, shell=True, text=True, stdout=log_file, stderr=log_file, cwd="MMVC_Trainer")
-            print("Training stated")
-            print(f"returncode:{train_proc.returncode}")
-    except Exception as e:
-        print("start training exception:", str(e))
-        return (ERROR,  str(e))
-
-    return (SUCCESS, "success")
-
-def stop_training():
-    global train_proc
-    if train_proc == None:
-        print("Training have not stated.")
-        return (ERROR, "Training have not stated.")
-
-    status = train_proc.poll()
-    if status != None:
-        print("Training have already ended.", status)
-        train_proc = None
-        return (ERROR, "Training have already ended. " + status)
-    else:
-        train_proc.kill()
-        print("Training have stoped.")
-        return (SUCCESS, "success")
-
-### Main
-def mod_post_pre_training(batch:int):
-    res = set_batch_size(batch)
-    if res[0] == ERROR:
-        return {"result":"failed", "detail": f"Preprocess(set_batch_size) failed. {res[1]}"}
-
-    res = set_dummy_device_count()
-    if res[0] == ERROR:
-        return {"result":"failed", "detail": f"Preprocess(set_dummy_device_count) failed. {res[1]}"}
-
-    res = create_dataset()
-    if res[0] == ERROR:
-        return {"result":"failed", "detail": f"Preprocess failed(create_dataset). {res[1]}"}
-
-    return {"result":"success", "detail": f"Preprocess succeeded. {res[1]}"}
-
-
-def mod_post_start_training(enable_finetuning:str, GModel:str, DModel:str):
-    print("START_TRAINING:::::::", enable_finetuning, GModel, DModel)
-    res = exec_training(enable_finetuning, GModel, DModel)
-    if res[0] == ERROR:
-        return {"result":"failed", "detail": f"Start training failed. {res[1]}"}
-
-    return {"result":"success", "detail": f"Start training succeeded. {res[1]}"}
-
-def mod_post_stop_training():
-    res = stop_training()
-    if res[0] == ERROR:
-        return {"result":"failed", "detail": f"Stop training failed. {res[1]}"}
-
-    return {"result":"success", "detail": f"Stop training succeeded. {res[1]}"}
-
-### DEBUG
-def mod_get_related_files():
-    files = get_file_list(os.path.join(LOG_DIR,"*"))
-    files.extend([
-        "MMVC_Trainer/dataset/multi_speaker_correspondence.txt",
-        "MMVC_Trainer/train_ms.py",
-    ])
-    files.extend(
-        get_file_list("MMVC_Trainer/configs/*")
-    )
-
-    res = []
-    for f in files:
-        size = os.path.getsize(f)
-        data = ""
-        if size < 1024*1024:
-            with open(f, "r") as input:
-                data = input.read()
-    
-        res.append({
-            "name":f,
-            "size":size,
-            "data":data
-        })
-
-    json_compatible_item_data = jsonable_encoder(res)
-    return JSONResponse(content=json_compatible_item_data)
-
-def mod_get_tail_training_log(num:int):
-    training_log_file = os.path.join(LOG_DIR, "training.txt")
-    res = sync_exec(f"cat {training_log_file} | sed -e 's/.*\r//' > /tmp/out","/dev/null")
-    cmd = f'tail -n {num} /tmp/out'
-    res = sync_exec_with_stdout(cmd, "/dev/null")
-    if res[0] == ERROR:
-        return {"result":"failed", "detail": f"Tail training log failed. {res[1]}"}
-    return {"result":"success", "detail":res[1]}

server/voice_changer/RVC/ModelSlotGenerator.py

@@ -1,126 +0,0 @@
-from const import EnumEmbedderTypes, EnumInferenceTypes
-
-import torch
-import onnxruntime
-import json
-
-from data.ModelSlot import ModelSlot
-
-
-def _setInfoByPytorch(slot: ModelSlot):
-    cpt = torch.load(slot.modelFile, map_location="cpu")
-    config_len = len(cpt["config"])
-
-    if config_len == 18:
-        # Original RVC
-        slot.f0 = True if cpt["f0"] == 1 else False
-        version = cpt.get("version", "v1")
-        if version is None or version == "v1":
-            slot.modelType = 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 = EnumInferenceTypes.pyTorchRVCv2.value if slot.f0 else EnumInferenceTypes.pyTorchRVCv2Nono.value
-            slot.embChannels = 768
-            slot.embOutputLayer = 12
-            slot.useFinalProj = False
-            slot.embedder = EnumEmbedderTypes.hubert.value
-            print("[Voice Changer] Official Model(pyTorch) : v2")
-
-    else:
-        # DDPN RVC
-        slot.f0 = True if cpt["f0"] == 1 else False
-        slot.modelType = EnumInferenceTypes.pyTorchWebUI.value if slot.f0 else EnumInferenceTypes.pyTorchWebUINono.value
-        slot.embChannels = cpt["config"][17]
-        slot.embOutputLayer = 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 slot.embChannels == 256 and slot.embOutputLayer == 9 and slot.useFinalProj is True:
-            print("[Voice Changer] DDPN Model(pyTorch) : Official v1 like")
-        elif slot.embChannels == 768 and slot.embOutputLayer == 12 and slot.useFinalProj is False:
-            print("[Voice Changer] DDPN Model(pyTorch): Official v2 like")
-        else:
-            print(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"):
-            slot.embedder = slot.embedder[:-3]
-
-        # if slot.embedder == EnumEmbedderTypes.hubert.value:
-        #     slot.embedder = EnumEmbedderTypes.hubert
-        # elif slot.embedder == EnumEmbedderTypes.contentvec.value:
-        #     slot.embedder = EnumEmbedderTypes.contentvec
-        # elif slot.embedder == EnumEmbedderTypes.hubert_jp.value:
-        #     slot.embedder = EnumEmbedderTypes.hubert_jp
-        # else:
-        #     raise RuntimeError("[Voice Changer][setInfoByONNX] unknown embedder")
-
-    slot.samplingRate = cpt["config"][-1]
-
-    del cpt
-
-
-def _setInfoByONNX(slot: ModelSlot):
-    tmp_onnx_session = onnxruntime.InferenceSession(slot.modelFile, providers=["CPUExecutionProvider"])
-    modelmeta = tmp_onnx_session.get_modelmeta()
-    try:
-        metadata = json.loads(modelmeta.custom_metadata_map["metadata"])
-
-        # slot.modelType = metadata["modelType"]
-        slot.embChannels = metadata["embChannels"]
-
-        slot.embOutputLayer = metadata["embOutputLayer"] if "embOutputLayer" in metadata else 9
-        slot.useFinalProj = metadata["useFinalProj"] if "useFinalProj" in metadata else True if slot.embChannels == 256 else False
-
-        if slot.embChannels == 256:
-            slot.useFinalProj = True
-        else:
-            slot.useFinalProj = False
-
-        # ONNXモデルの情報を表示
-        if slot.embChannels == 256 and slot.embOutputLayer == 9 and slot.useFinalProj is True:
-            print("[Voice Changer] ONNX Model: Official v1 like")
-        elif slot.embChannels == 768 and slot.embOutputLayer == 12 and slot.useFinalProj is False:
-            print("[Voice Changer] ONNX Model: Official v2 like")
-        else:
-            print(f"[Voice Changer] ONNX Model: ch:{slot.embChannels}, L:{slot.embOutputLayer}, FP:{slot.useFinalProj}")
-
-        if "embedder" not in metadata:
-            slot.embedder = EnumEmbedderTypes.hubert.value
-        else:
-            slot.embedder = metadata["embedder"]
-        # elif metadata["embedder"] == EnumEmbedderTypes.hubert.value:
-        #     slot.embedder = EnumEmbedderTypes.hubert
-        # elif metadata["embedder"] == EnumEmbedderTypes.contentvec.value:
-        #     slot.embedder = EnumEmbedderTypes.contentvec
-        # elif metadata["embedder"] == EnumEmbedderTypes.hubert_jp.value:
-        #     slot.embedder = EnumEmbedderTypes.hubert_jp
-        # else:
-        #     raise RuntimeError("[Voice Changer][setInfoByONNX] unknown embedder")
-
-        slot.f0 = metadata["f0"]
-        slot.modelType = EnumInferenceTypes.onnxRVC.value if slot.f0 else EnumInferenceTypes.onnxRVCNono.value
-        slot.samplingRate = metadata["samplingRate"]
-        slot.deprecated = False
-
-    except Exception as e:
-        slot.modelType = EnumInferenceTypes.onnxRVC.value
-        slot.embChannels = 256
-        slot.embedder = EnumEmbedderTypes.hubert.value
-        slot.f0 = True
-        slot.samplingRate = 48000
-        slot.deprecated = True
-
-        print("[Voice Changer] setInfoByONNX", e)
-        print("[Voice Changer] ############## !!!! CAUTION !!!! ####################")
-        print("[Voice Changer] This onnxfie is depricated. Please regenerate onnxfile.")
-        print("[Voice Changer] ############## !!!! CAUTION !!!! ####################")
-
-    del tmp_onnx_session

server/voice_changer/RVC/RVC.py

@@ -20,14 +20,10 @@ if sys.platform.startswith("darwin"):
 else:
     sys.path.append("RVC")
 
-from voice_changer.RVC.ModelSlotGenerator import (
+
-    _setInfoByONNX,
-    _setInfoByPytorch,
-)
 from voice_changer.RVC.RVCSettings import RVCSettings
 from voice_changer.RVC.embedder.EmbedderManager import EmbedderManager
-from voice_changer.utils.LoadModelParams import LoadModelParams2
+from voice_changer.utils.VoiceChangerModel import AudioInOut, VoiceChangerModel
-from voice_changer.utils.VoiceChangerModel import AudioInOut
 from voice_changer.utils.VoiceChangerParams import VoiceChangerParams
 from voice_changer.RVC.onnxExporter.export2onnx import export2onnx
 from voice_changer.RVC.pitchExtractor.PitchExtractorManager import PitchExtractorManager
@@ -35,10 +31,10 @@ from voice_changer.RVC.pipeline.PipelineGenerator import createPipeline
 from voice_changer.RVC.deviceManager.DeviceManager import DeviceManager
 from voice_changer.RVC.pipeline.Pipeline import Pipeline
 
-from Exceptions import DeviceCannotSupportHalfPrecisionException, NoModeLoadedException
+from Exceptions import DeviceCannotSupportHalfPrecisionException
 
 
-class RVC:
+class RVC(VoiceChangerModel):
     initialLoad: bool = True
     settings: RVCSettings = RVCSettings()
 
@@ -53,7 +49,7 @@ class RVC:
     needSwitch: bool = False
 
     def __init__(self, params: VoiceChangerParams, slotInfo: RVCModelSlot):
-        print("[Voice Changer][RVC] Creating instance ")
+        print("[Voice Changer] [RVC] Creating instance ")
         EmbedderManager.initialize(params)
 
         self.params = params
@@ -64,38 +60,16 @@ class RVC:
         self.initialize()
 
     def initialize(self):
-        print("[Voice Changer][RVC] Initializing... ")
+        print("[Voice Changer] [RVC] Initializing... ")
 
         # pipelineの生成
         self.pipeline = createPipeline(self.slotInfo, self.settings.gpu, self.settings.f0Detector)
 
         # その他の設定
-        self.trans = self.slotInfo.defaultTune
+        self.settings.tran = self.slotInfo.defaultTune
-        self.index_ratio = self.slotInfo.defaultIndexRatio
+        self.settings.indexRatio = self.slotInfo.defaultIndexRatio
-        self.protect = self.slotInfo.defaultProtect
+        self.settings.protect = self.slotInfo.defaultProtect
-        self.samplingRate = self.slotInfo.samplingRate
+        print("[Voice Changer] [RVC] Initializing... done")
-        print("[Voice Changer][RVC] Initializing... done")
-
-    @classmethod
-    def loadModel2(cls, props: LoadModelParams2):
-        slotInfo: RVCModelSlot = RVCModelSlot()
-        for file in props.files:
-            if file.kind == "rvcModel":
-                slotInfo.modelFile = file.name
-            elif file.kind == "rvcIndex":
-                slotInfo.indexFile = file.name
-        slotInfo.defaultTune = 0
-        slotInfo.defaultIndexRatio = 0
-        slotInfo.defaultProtect = 0.5
-        slotInfo.isONNX = slotInfo.modelFile.endswith(".onnx")
-        slotInfo.name = os.path.splitext(os.path.basename(slotInfo.modelFile))[0]
-        # slotInfo.iconFile = "/assets/icons/noimage.png"
-
-        if slotInfo.isONNX:
-            _setInfoByONNX(slotInfo)
-        else:
-            _setInfoByPytorch(slotInfo)
-        return slotInfo
 
     def update_settings(self, key: str, val: int | float | str):
         print("[Voice Changer][RVC]: update_settings", key, val)
@@ -123,7 +97,7 @@ class RVC:
         return data
 
     def get_processing_sampling_rate(self):
-        return self.settings.modelSamplingRate
+        return self.slotInfo.samplingRate
 
     def generate_input(
         self,
@@ -170,14 +144,6 @@ class RVC:
         return (audio_buffer, convertSize, vol)
 
     def inference(self, data):
-        # if self.settings.modelSlotIndex < 0:
-        #     print(
-        #         "[Voice Changer] wait for loading model...",
-        #         self.settings.modelSlotIndex,
-        #         self.currentSlot,
-        #     )
-        #     raise NoModeLoadedException("model_common")
-
         audio = data[0]
         convertSize = data[1]
         vol = data[2]
@@ -185,17 +151,13 @@ class RVC:
         if vol < self.settings.silentThreshold:
             return np.zeros(convertSize).astype(np.int16)
 
-        audio = torchaudio.functional.resample(audio, self.settings.modelSamplingRate, 16000, rolloff=0.99)
+        audio = torchaudio.functional.resample(audio, self.slotInfo.samplingRate, 16000, rolloff=0.99)
         repeat = 1 if self.settings.rvcQuality else 0
         sid = 0
         f0_up_key = self.settings.tran
         index_rate = self.settings.indexRatio
         protect = self.settings.protect
 
-        # if_f0 = 1 if self.settings.modelSlots[self.currentSlot].f0 else 0
-        # embOutputLayer = self.settings.modelSlots[self.currentSlot].embOutputLayer
-        # useFinalProj = self.settings.modelSlots[self.currentSlot].useFinalProj
-
         if_f0 = 1 if self.slotInfo.f0 else 0
         embOutputLayer = self.slotInfo.embOutputLayer
         useFinalProj = self.slotInfo.useFinalProj
@@ -207,7 +169,7 @@ class RVC:
                 f0_up_key,
                 index_rate,
                 if_f0,
-                self.settings.extraConvertSize / self.settings.modelSamplingRate,  # extaraDataSizeの秒数。RVCのモデルのサンプリングレートで処理(★１)。
+                self.settings.extraConvertSize / self.slotInfo.samplingRate,  # extaraDataSizeの秒数。RVCのモデルのサンプリングレートで処理(★１)。
                 embOutputLayer,
                 useFinalProj,
                 repeat,
@@ -292,36 +254,6 @@ class RVC:
         # self.settings.modelSlotIndex = targetSlot
         # self.currentSlot = self.settings.modelSlotIndex
 
-    # def update_model_default(self):
-    #     # {"slot":9,"key":"name","val":"dogsdododg"}
-    #     self.modelSlotManager.update_model_info(
-    #         json.dumps(
-    #             {
-    #                 "slot": self.currentSlot,
-    #                 "key": "defaultTune",
-    #                 "val": self.settings.tran,
-    #             }
-    #         )
-    #     )
-    #     self.modelSlotManager.update_model_info(
-    #         json.dumps(
-    #             {
-    #                 "slot": self.currentSlot,
-    #                 "key": "defaultIndexRatio",
-    #                 "val": self.settings.indexRatio,
-    #             }
-    #         )
-    #     )
-    #     self.modelSlotManager.update_model_info(
-    #         json.dumps(
-    #             {
-    #                 "slot": self.currentSlot,
-    #                 "key": "defaultProtect",
-    #                 "val": self.settings.protect,
-    #             }
-    #         )
-    #     )
-
     def get_model_current(self):
         return [
             {
@@ -337,9 +269,3 @@ class RVC:
                 "val": self.settings.protect,
             },
         ]
-
-    # def update_model_info(self, newData: str):
-    #     self.modelSlotManager.update_model_info(newData)
-
-    # def upload_model_assets(self, params: str):
-    #     self.modelSlotManager.store_model_assets(params)

server/voice_changer/RVC/RVCModelSlotGenerator.py

@@ -0,0 +1,152 @@
+import os
+from const import EnumEmbedderTypes, EnumInferenceTypes
+
+import torch
+import onnxruntime
+import json
+
+from data.ModelSlot import ModelSlot, RVCModelSlot
+from voice_changer.utils.LoadModelParams import LoadModelParams
+from voice_changer.utils.ModelSlotGenerator import ModelSlotGenerator
+
+
+class RVCModelSlotGenerator(ModelSlotGenerator):
+    @classmethod
+    def loadModel(cls, props: LoadModelParams):
+        slotInfo: RVCModelSlot = RVCModelSlot()
+        for file in props.files:
+            if file.kind == "rvcModel":
+                slotInfo.modelFile = file.name
+            elif file.kind == "rvcIndex":
+                slotInfo.indexFile = file.name
+        slotInfo.defaultTune = 0
+        slotInfo.defaultIndexRatio = 0
+        slotInfo.defaultProtect = 0.5
+        slotInfo.isONNX = slotInfo.modelFile.endswith(".onnx")
+        slotInfo.name = os.path.splitext(os.path.basename(slotInfo.modelFile))[0]
+        # slotInfo.iconFile = "/assets/icons/noimage.png"
+
+        if slotInfo.isONNX:
+            cls._setInfoByONNX(slotInfo)
+        else:
+            cls._setInfoByPytorch(slotInfo)
+        return slotInfo
+
+    @classmethod
+    def _setInfoByPytorch(cls, slot: ModelSlot):
+        cpt = torch.load(slot.modelFile, map_location="cpu")
+        config_len = len(cpt["config"])
+
+        if config_len == 18:
+            # Original RVC
+            slot.f0 = True if cpt["f0"] == 1 else False
+            version = cpt.get("version", "v1")
+            if version is None or version == "v1":
+                slot.modelType = 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 = EnumInferenceTypes.pyTorchRVCv2.value if slot.f0 else EnumInferenceTypes.pyTorchRVCv2Nono.value
+                slot.embChannels = 768
+                slot.embOutputLayer = 12
+                slot.useFinalProj = False
+                slot.embedder = EnumEmbedderTypes.hubert.value
+                print("[Voice Changer] Official Model(pyTorch) : v2")
+
+        else:
+            # DDPN RVC
+            slot.f0 = True if cpt["f0"] == 1 else False
+            slot.modelType = EnumInferenceTypes.pyTorchWebUI.value if slot.f0 else EnumInferenceTypes.pyTorchWebUINono.value
+            slot.embChannels = cpt["config"][17]
+            slot.embOutputLayer = 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 slot.embChannels == 256 and slot.embOutputLayer == 9 and slot.useFinalProj is True:
+                print("[Voice Changer] DDPN Model(pyTorch) : Official v1 like")
+            elif slot.embChannels == 768 and slot.embOutputLayer == 12 and slot.useFinalProj is False:
+                print("[Voice Changer] DDPN Model(pyTorch): Official v2 like")
+            else:
+                print(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"):
+                slot.embedder = slot.embedder[:-3]
+
+            # if slot.embedder == EnumEmbedderTypes.hubert.value:
+            #     slot.embedder = EnumEmbedderTypes.hubert
+            # elif slot.embedder == EnumEmbedderTypes.contentvec.value:
+            #     slot.embedder = EnumEmbedderTypes.contentvec
+            # elif slot.embedder == EnumEmbedderTypes.hubert_jp.value:
+            #     slot.embedder = EnumEmbedderTypes.hubert_jp
+            # else:
+            #     raise RuntimeError("[Voice Changer][setInfoByONNX] unknown embedder")
+
+        slot.samplingRate = cpt["config"][-1]
+
+        del cpt
+
+    @classmethod
+    def _setInfoByONNX(cls, slot: ModelSlot):
+        tmp_onnx_session = onnxruntime.InferenceSession(slot.modelFile, providers=["CPUExecutionProvider"])
+        modelmeta = tmp_onnx_session.get_modelmeta()
+        try:
+            metadata = json.loads(modelmeta.custom_metadata_map["metadata"])
+
+            # slot.modelType = metadata["modelType"]
+            slot.embChannels = metadata["embChannels"]
+
+            slot.embOutputLayer = metadata["embOutputLayer"] if "embOutputLayer" in metadata else 9
+            slot.useFinalProj = metadata["useFinalProj"] if "useFinalProj" in metadata else True if slot.embChannels == 256 else False
+
+            if slot.embChannels == 256:
+                slot.useFinalProj = True
+            else:
+                slot.useFinalProj = False
+
+            # ONNXモデルの情報を表示
+            if slot.embChannels == 256 and slot.embOutputLayer == 9 and slot.useFinalProj is True:
+                print("[Voice Changer] ONNX Model: Official v1 like")
+            elif slot.embChannels == 768 and slot.embOutputLayer == 12 and slot.useFinalProj is False:
+                print("[Voice Changer] ONNX Model: Official v2 like")
+            else:
+                print(f"[Voice Changer] ONNX Model: ch:{slot.embChannels}, L:{slot.embOutputLayer}, FP:{slot.useFinalProj}")
+
+            if "embedder" not in metadata:
+                slot.embedder = EnumEmbedderTypes.hubert.value
+            else:
+                slot.embedder = metadata["embedder"]
+            # elif metadata["embedder"] == EnumEmbedderTypes.hubert.value:
+            #     slot.embedder = EnumEmbedderTypes.hubert
+            # elif metadata["embedder"] == EnumEmbedderTypes.contentvec.value:
+            #     slot.embedder = EnumEmbedderTypes.contentvec
+            # elif metadata["embedder"] == EnumEmbedderTypes.hubert_jp.value:
+            #     slot.embedder = EnumEmbedderTypes.hubert_jp
+            # else:
+            #     raise RuntimeError("[Voice Changer][setInfoByONNX] unknown embedder")
+
+            slot.f0 = metadata["f0"]
+            slot.modelType = EnumInferenceTypes.onnxRVC.value if slot.f0 else EnumInferenceTypes.onnxRVCNono.value
+            slot.samplingRate = metadata["samplingRate"]
+            slot.deprecated = False
+
+        except Exception as e:
+            slot.modelType = EnumInferenceTypes.onnxRVC.value
+            slot.embChannels = 256
+            slot.embedder = EnumEmbedderTypes.hubert.value
+            slot.f0 = True
+            slot.samplingRate = 48000
+            slot.deprecated = True
+
+            print("[Voice Changer] setInfoByONNX", e)
+            print("[Voice Changer] ############## !!!! CAUTION !!!! ####################")
+            print("[Voice Changer] This onnxfie is depricated. Please regenerate onnxfile.")
+            print("[Voice Changer] ############## !!!! CAUTION !!!! ####################")
+
+        del tmp_onnx_session

server/voice_changer/RVC/RVCSettings.py

@@ -7,12 +7,9 @@ class RVCSettings:
     dstId: int = 0
 
     f0Detector: str = "harvest"  # dio or harvest
-    tran: int = 20
+    tran: int = 12
     silentThreshold: float = 0.00001
-    extraConvertSize: int = 1024 * 32
+    extraConvertSize: int = 1024 * 4
-    clusterInferRatio: float = 0.1
-
-    framework: str = "PyTorch"  # PyTorch or ONNX
 
     indexRatio: float = 0
     protect: float = 0.5
@@ -21,8 +18,8 @@ class RVCSettings:
     modelSamplingRate: int = 48000
 
     speakers: dict[str, int] = field(default_factory=lambda: {})
-    isHalf: int = 1  # 0:off, 1:on
+    # isHalf: int = 1  # 0:off, 1:on
-    enableDirectML: int = 0  # 0:off, 1:on
+    # enableDirectML: int = 0  # 0:off, 1:on
     # ↓mutableな物だけ列挙
     intData = [
         "gpu",
@@ -30,11 +27,7 @@ class RVCSettings:
         "tran",
         "extraConvertSize",
         "rvcQuality",
-        "modelSamplingRate",
         "silenceFront",
-        "modelSlotIndex",
-        "isHalf",
-        "enableDirectML",
     ]
     floatData = ["silentThreshold", "indexRatio", "protect"]
-    strData = ["framework", "f0Detector"]
+    strData = ["f0Detector"]

server/voice_changer/VoiceChangerManager.py

@@ -7,7 +7,7 @@ from voice_changer.Local.ServerDevice import ServerDevice, ServerDeviceCallbacks
 from voice_changer.ModelSlotManager import ModelSlotManager
 from voice_changer.VoiceChanger import VoiceChanger
 from const import UPLOAD_DIR, ModelType
-from voice_changer.utils.LoadModelParams import LoadModelParamFile, LoadModelParams, LoadModelParams2
+from voice_changer.utils.LoadModelParams import LoadModelParams
 from voice_changer.utils.VoiceChangerModel import AudioInOut
 from voice_changer.utils.VoiceChangerParams import VoiceChangerParams
 from dataclasses import dataclass, asdict, field
@@ -85,20 +85,15 @@ class VoiceChangerManager(ServerDeviceCallbacks):
             cls._instance.voiceChanger = VoiceChanger(params)
         return cls._instance
 
-    def loadModel(self, props: LoadModelParams):
+    def loadModel(self, params: LoadModelParams):
-        paramDict = props.params
+        if params.isSampleMode:
-        if paramDict["sampleId"] is not None:
             # サンプルダウンロード
-            downloadSample(self.params.sample_mode, paramDict["sampleId"], self.params.model_dir, props.slot, {"useIndex": paramDict["rvcIndexDownload"]})
+            downloadSample(self.params.sample_mode, params.sampleId, self.params.model_dir, params.slot, {"useIndex": params.params["rvcIndexDownload"]})
             self.modelSlotManager.getAllSlotInfo(reload=True)
             info = {"status": "OK"}
             return info
-        elif paramDict["voiceChangerType"]:
+        else:
-            # 新しいアップローダ
+            # アップローダ
-            # Dataを展開
-            params = LoadModelParams2(**paramDict)
-            params.files = [LoadModelParamFile(**x) for x in paramDict["files"]]
-
             # ファイルをslotにコピー
             for file in params.files:
                 print("FILE", file)
@@ -116,42 +111,32 @@ class VoiceChangerManager(ServerDeviceCallbacks):
 
             # メタデータ作成(各VCで定義)
             if params.voiceChangerType == "RVC":
-                from voice_changer.RVC.RVC import RVC  # 起動時にインポートするとパラメータが取れない。
+                from voice_changer.RVC.RVCModelSlotGenerator import RVCModelSlotGenerator  # 起動時にインポートするとパラメータが取れない。
 
-                slotInfo = RVC.loadModel2(params)
+                slotInfo = RVCModelSlotGenerator.loadModel(params)
                 self.modelSlotManager.save_model_slot(params.slot, slotInfo)
             elif params.voiceChangerType == "MMVCv13":
                 from voice_changer.MMVCv13.MMVCv13 import MMVCv13
 
-                slotInfo = MMVCv13.loadModel2(params)
+                slotInfo = MMVCv13.loadModel(params)
                 self.modelSlotManager.save_model_slot(params.slot, slotInfo)
             elif params.voiceChangerType == "MMVCv15":
                 from voice_changer.MMVCv15.MMVCv15 import MMVCv15
 
-                slotInfo = MMVCv15.loadModel2(params)
+                slotInfo = MMVCv15.loadModel(params)
                 self.modelSlotManager.save_model_slot(params.slot, slotInfo)
             elif params.voiceChangerType == "so-vits-svc-40":
                 from voice_changer.SoVitsSvc40.SoVitsSvc40 import SoVitsSvc40
 
-                slotInfo = SoVitsSvc40.loadModel2(params)
+                slotInfo = SoVitsSvc40.loadModel(params)
                 self.modelSlotManager.save_model_slot(params.slot, slotInfo)
             elif params.voiceChangerType == "DDSP-SVC":
                 from voice_changer.DDSP_SVC.DDSP_SVC import DDSP_SVC
 
-                slotInfo = DDSP_SVC.loadModel2(params)
+                slotInfo = DDSP_SVC.loadModel(params)
                 self.modelSlotManager.save_model_slot(params.slot, slotInfo)
             print("params", params)
 
-        else:
-            # 古いアップローダ
-            print("[Voice Canger]: upload models........")
-            info = self.voiceChanger.loadModel(props)
-            if hasattr(info, "status") and info["status"] == "NG":
-                return info
-            else:
-                info["status"] = "OK"
-                return info
-
     def get_info(self):
         data = asdict(self.settings)
         data["gpus"] = self.gpus

server/voice_changer/utils/LoadModelParams.py

@@ -1,27 +1,8 @@
 from dataclasses import dataclass
-from typing import Any
 
 from const import VoiceChangerType
 from typing import Literal, TypeAlias
 
-
-@dataclass
-class FilePaths:
-    configFilename: str | None
-    pyTorchModelFilename: str | None
-    onnxModelFilename: str | None
-    clusterTorchModelFilename: str | None
-    featureFilename: str | None
-    indexFilename: str | None
-
-
-@dataclass
-class LoadModelParams:
-    slot: int
-    isHalf: bool
-    params: Any
-
-
 LoadModelParamFileKind: TypeAlias = Literal[
     "mmvcv13Config",
     "mmvcv13Model",
@@ -47,9 +28,10 @@ class LoadModelParamFile:
 
 
 @dataclass
-class LoadModelParams2:
+class LoadModelParams:
     voiceChangerType: VoiceChangerType
     slot: int
     isSampleMode: bool
     sampleId: str
     files: list[LoadModelParamFile]
+    params: dict

server/voice_changer/utils/ModelSlotGenerator.py

@@ -0,0 +1,9 @@
+from typing import Protocol
+
+from voice_changer.utils.LoadModelParams import LoadModelParams
+
+
+class ModelSlotGenerator(Protocol):
+    @classmethod
+    def loadModel(self, params: LoadModelParams):
+        ...

server/voice_changer/utils/VoiceChangerModel.py

@@ -21,10 +21,8 @@ class VoiceChangerModel(Protocol):
     def inference(self, data: tuple[Any, ...]) -> Any:
         ...
 
-    def generate_input(
+    def generate_input(self, newData: AudioInOut, inputSize: int, crossfadeSize: int, solaSearchFrame: int) -> tuple[Any, ...]:
-        self, newData: AudioInOut, inputSize: int, crossfadeSize: int
-    ) -> tuple[Any, ...]:
         ...
 
-    def update_settings(self, key: str, val: Any) -> bool:
+    def update_settings(self, key: str, val: int | float | str) -> bool:
         ...