PublicMirrors/voice-changer
1
0
Fork 0
mirror of https://github.com/w-okada/voice-changer.git synced 2025-02-02 16:23:58 +03:00
Code Issues Actions Packages Projects Releases Wiki Activity

remove slotindex from json

Browse Source
This commit is contained in:
w-okada 2023-08-05 13:24:11 +09:00
parent db97441380
commit 6d4c138821
4 changed files with 19 additions and 53 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

4
server/voice_changer/DiffusionSVC/DiffusionSVC.py
View File

@ -215,6 +215,10 @@ class DiffusionSVC(VoiceChangerModel):
"key": "defaultTune",
"val": self.settings.tran,
},
{
"key": "dstId",
"val": self.settings.dstId,
},
{
"key": "defaultKstep",
"val": self.settings.kStep,

57
server/voice_changer/DiffusionSVC/DiffusionSVCModelSlotGenerator.py
View File

@ -6,9 +6,11 @@ import json
from data.ModelSlot import DiffusionSVCModelSlot, ModelSlot, RVCModelSlot
from voice_changer.DiffusionSVC.inferencer.diffusion_svc_model.diffusion.unit2mel import load_model_vocoder_from_combo
from voice_changer.VoiceChangerParamsManager import VoiceChangerParamsManager
from voice_changer.utils.LoadModelParams import LoadModelParams
from voice_changer.utils.ModelSlotGenerator import ModelSlotGenerator
def get_divisors(n):
divisors = []
for i in range(1, int(n**0.5)+1):
@ -31,6 +33,7 @@ class DiffusionSVCModelSlotGenerator(ModelSlotGenerator):
slotInfo.name = os.path.splitext(os.path.basename(slotInfo.modelFile))[0]
# slotInfo.iconFile = "/assets/icons/noimage.png"
slotInfo.embChannels = 768
slotInfo.slotIndex = props.slot
if slotInfo.isONNX:
slotInfo = cls._setInfoByONNX(slotInfo)
@ -40,7 +43,10 @@ class DiffusionSVCModelSlotGenerator(ModelSlotGenerator):
@classmethod
def _setInfoByPytorch(cls, slot: DiffusionSVCModelSlot):
diff_model, diff_args, naive_model, naive_args = load_model_vocoder_from_combo(slot.modelFile, device="cpu")
vcparams = VoiceChangerParamsManager.get_instance().params
modelPath = os.path.join(vcparams.model_dir, str(slot.slotIndex), os.path.basename(slot.modelFile))
diff_model, diff_args, naive_model, naive_args = load_model_vocoder_from_combo(modelPath, device="cpu")
slot.kStepMax = diff_args.model.k_step_max
slot.nLayers = diff_args.model.n_layers
slot.nnLayers = naive_args.model.n_layers
@ -52,53 +58,4 @@ class DiffusionSVCModelSlotGenerator(ModelSlotGenerator):
@classmethod
def _setInfoByONNX(cls, slot: ModelSlot):
tmp_onnx_session = onnxruntime.InferenceSession(slot.modelFile, providers=["CPUExecutionProvider"])
modelmeta = tmp_onnx_session.get_modelmeta()
try:
slot = RVCModelSlot(**asdict(slot))
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 = "hubert_base"
else:
slot.embedder = metadata["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 = "hubert_base"
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
return slot

1
server/voice_changer/DiffusionSVC/DiffusionSVCSettings.py
View File

@ -13,6 +13,7 @@ class DiffusionSVCSettings:
kStep: int = 20
speedUp: int = 10
skipDiffusion: int = 0 # 0:off, 1:on
silenceFront: int = 1 # 0:off, 1:on
modelSamplingRate: int = 44100

10
server/voice_changer/DiffusionSVC/pipeline/PipelineGenerator.py
View File

@ -7,19 +7,23 @@ from voice_changer.DiffusionSVC.pitchExtractor.PitchExtractorManager import Pitc
from voice_changer.RVC.deviceManager.DeviceManager import DeviceManager
from voice_changer.RVC.embedder.EmbedderManager import EmbedderManager
import os
import torch
from torchaudio.transforms import Resample
from voice_changer.VoiceChangerParamsManager import VoiceChangerParamsManager
def createPipeline(modelSlot: DiffusionSVCModelSlot, gpu: int, f0Detector: str, inputSampleRate: int, outputSampleRate: int):
dev = DeviceManager.get_instance().getDevice(gpu)
vcparams = VoiceChangerParamsManager.get_instance().params
# half = DeviceManager.get_instance().halfPrecisionAvailable(gpu)
half = False
# Inferencer 生成
try:
inferencer = InferencerManager.getInferencer(modelSlot.modelType, modelSlot.modelFile, gpu)
try:
modelPath = os.path.join(vcparams.model_dir, str(modelSlot.slotIndex), os.path.basename(modelSlot.modelFile))
inferencer = InferencerManager.getInferencer(modelSlot.modelType, modelPath, gpu)
except Exception as e:
print("[Voice Changer] exception! loading inferencer", e)
traceback.print_exc()