mirror of
https://github.com/w-okada/voice-changer.git
synced 2025-02-02 16:23:58 +03:00
WIP:refactoring
This commit is contained in:
wataru
parent
be42bb682d
commit
9806ce2f3d
@ -15,7 +15,11 @@ class ServerAudioDevice:
|
||||
|
||||
|
||||
def list_audio_device():
|
||||
try:
|
||||
audioDeviceList = sd.query_devices()
|
||||
except Exception as e:
|
||||
print("[Voice Changer] ex: query_devices", e)
|
||||
return [], []
|
||||
|
||||
inputAudioDeviceList = [d for d in audioDeviceList if d["max_input_channels"] > 0]
|
||||
outputAudioDeviceList = [d for d in audioDeviceList if d["max_output_channels"] > 0]
|
||||
|
||||
@ -4,34 +4,66 @@ import numpy as np
|
||||
import librosa
|
||||
import sounddevice as sd
|
||||
|
||||
from dataclasses import dataclass, asdict, field
|
||||
|
||||
from voice_changer.Local.AudioDeviceList import ServerAudioDevice
|
||||
from voice_changer.VoiceChanger import VoiceChanger
|
||||
from voice_changer.VoiceChangerManager import VoiceChangerManager
|
||||
from voice_changer.utils.Timer import Timer
|
||||
|
||||
|
||||
@dataclass()
|
||||
class ServerDeviceSettings:
|
||||
enableServerAudio: int = 0 # 0:off, 1:on
|
||||
serverAudioStated: int = 0 # 0:off, 1:on
|
||||
serverInputAudioSampleRate: int = 44100
|
||||
serverOutputAudioSampleRate: int = 44100
|
||||
serverInputDeviceId: int = -1
|
||||
serverOutputDeviceId: int = -1
|
||||
serverReadChunkSize: int = 256
|
||||
serverInputAudioGain: float = 1.0
|
||||
serverOutputAudioGain: float = 1.0
|
||||
|
||||
|
||||
EditableServerDeviceSettings = {
|
||||
"intData": [
|
||||
"enableServerAudio",
|
||||
"serverAudioStated",
|
||||
"serverInputAudioSampleRate",
|
||||
"serverOutputAudioSampleRate",
|
||||
"serverInputDeviceId",
|
||||
"serverOutputDeviceId",
|
||||
"serverReadChunkSize",
|
||||
],
|
||||
"floatData": [
|
||||
"serverInputAudioGain",
|
||||
"serverOutputAudioGain",
|
||||
],
|
||||
}
|
||||
|
||||
|
||||
class ServerDevice:
|
||||
def __init__(self):
|
||||
self.voiceChanger: VoiceChanger | None = None
|
||||
pass
|
||||
def __init__(self, voiceChangerManager: VoiceChangerManager):
|
||||
self.settings = ServerDeviceSettings()
|
||||
self.voiceChangerManager: VoiceChangerManager = voiceChangerManager
|
||||
|
||||
def audio_callback(self, indata: np.ndarray, outdata: np.ndarray, frames, times, status):
|
||||
if self.voiceChanger is None:
|
||||
if self.voiceChangerManager.voiceChanger is None:
|
||||
print("[Voice Changer] voiceChanger is None")
|
||||
return
|
||||
|
||||
try:
|
||||
indata = indata * self.voiceChanger.settings.serverInputAudioGain
|
||||
indata = indata * self.settings.serverInputAudioGain
|
||||
with Timer("all_inference_time") as t:
|
||||
unpackedData = librosa.to_mono(indata.T) * 32768.0
|
||||
out_wav, times = self.voiceChanger.on_request(unpackedData)
|
||||
out_wav, times = self.voiceChangerManager.voiceChanger.on_request(unpackedData)
|
||||
outputChunnels = outdata.shape[1]
|
||||
outdata[:] = np.repeat(out_wav, outputChunnels).reshape(-1, outputChunnels) / 32768.0
|
||||
outdata[:] = outdata * self.voiceChanger.settings.serverOutputAudioGain
|
||||
outdata[:] = outdata * self.settings.serverOutputAudioGain
|
||||
all_inference_time = t.secs
|
||||
performance = [all_inference_time] + times
|
||||
if self.voiceChanger.emitTo is not None:
|
||||
self.voiceChanger.emitTo(performance)
|
||||
self.voiceChanger.settings.performance = [round(x * 1000) for x in performance]
|
||||
if self.voiceChangerManager.voiceChanger.emitTo is not None:
|
||||
self.voiceChangerManager.voiceChanger.emitTo(performance)
|
||||
self.voiceChangerManager.voiceChanger.settings.performance = [round(x * 1000) for x in performance]
|
||||
except Exception as e:
|
||||
print("[Voice Changer] ex:", e)
|
||||
|
||||
@ -42,31 +74,26 @@ class ServerDevice:
|
||||
else:
|
||||
return None
|
||||
|
||||
def serverLocal(self, _vc: VoiceChanger):
|
||||
self.voiceChanger = _vc
|
||||
vc = self.voiceChanger
|
||||
|
||||
def serverLocal(self):
|
||||
currentInputDeviceId = -1
|
||||
currentModelSamplingRate = -1
|
||||
currentOutputDeviceId = -1
|
||||
currentInputChunkNum = -1
|
||||
while True:
|
||||
if vc.settings.serverAudioStated == 0 or vc.settings.serverInputDeviceId == -1 or vc is None:
|
||||
vc.settings.inputSampleRate = 48000
|
||||
if self.settings.serverAudioStated == 0 or self.settings.serverInputDeviceId == -1 or self.voiceChangerManager is None:
|
||||
self.voiceChangerManager.voiceChanger.settings.inputSampleRate = 48000
|
||||
time.sleep(2)
|
||||
else:
|
||||
sd._terminate()
|
||||
sd._initialize()
|
||||
|
||||
sd.default.device[0] = vc.settings.serverInputDeviceId
|
||||
currentInputDeviceId = vc.settings.serverInputDeviceId
|
||||
sd.default.device[1] = vc.settings.serverOutputDeviceId
|
||||
currentOutputDeviceId = vc.settings.serverOutputDeviceId
|
||||
sd.default.device[0] = self.settings.serverInputDeviceId
|
||||
currentInputDeviceId = self.settings.serverInputDeviceId
|
||||
sd.default.device[1] = self.settings.serverOutputDeviceId
|
||||
currentOutputDeviceId = self.settings.serverOutputDeviceId
|
||||
|
||||
currentInputChannelNum = vc.settings.serverAudioInputDevices
|
||||
|
||||
serverInputAudioDevice = self.getServerAudioDevice(vc.settings.serverAudioInputDevices, currentInputDeviceId)
|
||||
serverOutputAudioDevice = self.getServerAudioDevice(vc.settings.serverAudioOutputDevices, currentOutputDeviceId)
|
||||
serverInputAudioDevice = self.getServerAudioDevice(self.voiceChangerManager.serverAudioInputDevices, currentInputDeviceId)
|
||||
serverOutputAudioDevice = self.getServerAudioDevice(self.voiceChangerManager.serverAudioOutputDevices, currentOutputDeviceId)
|
||||
print(serverInputAudioDevice, serverOutputAudioDevice)
|
||||
if serverInputAudioDevice is None or serverOutputAudioDevice is None:
|
||||
time.sleep(2)
|
||||
@ -76,12 +103,12 @@ class ServerDevice:
|
||||
currentInputChannelNum = serverInputAudioDevice.maxInputChannels
|
||||
currentOutputChannelNum = serverOutputAudioDevice.maxOutputChannels
|
||||
|
||||
currentInputChunkNum = vc.settings.serverReadChunkSize
|
||||
currentInputChunkNum = self.settings.serverReadChunkSize
|
||||
block_frame = currentInputChunkNum * 128
|
||||
|
||||
# sample rate precheck(alsa cannot use 40000?)
|
||||
try:
|
||||
currentModelSamplingRate = self.voiceChanger.voiceChangerModel.get_processing_sampling_rate()
|
||||
currentModelSamplingRate = self.voiceChangerManager.voiceChanger.voiceChangerModel.get_processing_sampling_rate()
|
||||
except Exception as e:
|
||||
print("[Voice Changer] ex: get_processing_sampling_rate", e)
|
||||
continue
|
||||
@ -94,38 +121,38 @@ class ServerDevice:
|
||||
channels=[currentInputChannelNum, currentOutputChannelNum],
|
||||
):
|
||||
pass
|
||||
vc.settings.serverInputAudioSampleRate = currentModelSamplingRate
|
||||
vc.settings.inputSampleRate = currentModelSamplingRate
|
||||
print(f"[Voice Changer] sample rate {vc.settings.serverInputAudioSampleRate}")
|
||||
self.settings.serverInputAudioSampleRate = currentModelSamplingRate
|
||||
self.voiceChangerManager.voiceChanger.settings.inputSampleRate = currentModelSamplingRate
|
||||
print(f"[Voice Changer] sample rate {self.settings.serverInputAudioSampleRate}")
|
||||
except Exception as e:
|
||||
print(
|
||||
"[Voice Changer] ex: fallback to device default samplerate",
|
||||
e,
|
||||
)
|
||||
vc.settings.serverInputAudioSampleRate = serverInputAudioDevice.default_samplerate
|
||||
vc.settings.inputSampleRate = vc.settings.serverInputAudioSampleRate
|
||||
self.settings.serverInputAudioSampleRate = serverInputAudioDevice.default_samplerate
|
||||
self.voiceChangerManager.voiceChanger.settings.inputSampleRate = self.settings.serverInputAudioSampleRate
|
||||
|
||||
# main loop
|
||||
try:
|
||||
with sd.Stream(
|
||||
callback=self.audio_callback,
|
||||
blocksize=block_frame,
|
||||
samplerate=vc.settings.serverInputAudioSampleRate,
|
||||
samplerate=self.settings.serverInputAudioSampleRate,
|
||||
dtype="float32",
|
||||
channels=[currentInputChannelNum, currentOutputChannelNum],
|
||||
):
|
||||
while vc.settings.serverAudioStated == 1 and currentInputDeviceId == vc.settings.serverInputDeviceId and currentOutputDeviceId == vc.settings.serverOutputDeviceId and currentModelSamplingRate == self.voiceChanger.voiceChangerModel.get_processing_sampling_rate() and currentInputChunkNum == vc.settings.serverReadChunkSize:
|
||||
while self.settings.serverAudioStated == 1 and currentInputDeviceId == self.settings.serverInputDeviceId and currentOutputDeviceId == self.settings.serverOutputDeviceId and currentModelSamplingRate == self.voiceChangerManager.voiceChanger.voiceChangerModel.get_processing_sampling_rate() and currentInputChunkNum == self.settings.serverReadChunkSize:
|
||||
time.sleep(2)
|
||||
print(
|
||||
"[Voice Changer] server audio",
|
||||
vc.settings.performance,
|
||||
self.voiceChangerManager.settings.performance,
|
||||
)
|
||||
print(
|
||||
"[Voice Changer] info:",
|
||||
vc.settings.serverAudioStated,
|
||||
self.settings.serverAudioStated,
|
||||
currentInputDeviceId,
|
||||
currentOutputDeviceId,
|
||||
vc.settings.serverInputAudioSampleRate,
|
||||
self.settings.serverInputAudioSampleRate,
|
||||
currentInputChunkNum,
|
||||
)
|
||||
|
||||
|
||||
@ -51,22 +51,7 @@ class VoiceChangerSettings:
|
||||
crossFadeOverlapSize: int = 4096
|
||||
|
||||
recordIO: int = 0 # 0:off, 1:on
|
||||
serverAudioInputDevices: list[ServerAudioDevice] = field(default_factory=lambda: [])
|
||||
serverAudioOutputDevices: list[ServerAudioDevice] = field(default_factory=lambda: [])
|
||||
|
||||
enableServerAudio: int = 0 # 0:off, 1:on
|
||||
serverAudioStated: int = 0 # 0:off, 1:on
|
||||
# serverInputAudioSampleRate: int = 48000
|
||||
# serverOutputAudioSampleRate: int = 48000
|
||||
serverInputAudioSampleRate: int = 44100
|
||||
serverOutputAudioSampleRate: int = 44100
|
||||
# serverInputAudioBufferSize: int = 1024 * 24
|
||||
# serverOutputAudioBufferSize: int = 1024 * 24
|
||||
serverInputDeviceId: int = -1
|
||||
serverOutputDeviceId: int = -1
|
||||
serverReadChunkSize: int = 256
|
||||
serverInputAudioGain: float = 1.0
|
||||
serverOutputAudioGain: float = 1.0
|
||||
performance: list[int] = field(default_factory=lambda: [0, 0, 0, 0])
|
||||
|
||||
# ↓mutableな物だけ列挙
|
||||
@ -75,23 +60,12 @@ class VoiceChangerSettings:
|
||||
"inputSampleRate",
|
||||
"crossFadeOverlapSize",
|
||||
"recordIO",
|
||||
"enableServerAudio",
|
||||
"serverAudioStated",
|
||||
"serverInputAudioSampleRate",
|
||||
"serverOutputAudioSampleRate",
|
||||
# "serverInputAudioBufferSize",
|
||||
# "serverOutputAudioBufferSize",
|
||||
"serverInputDeviceId",
|
||||
"serverOutputDeviceId",
|
||||
"serverReadChunkSize",
|
||||
]
|
||||
)
|
||||
floatData: list[str] = field(
|
||||
default_factory=lambda: [
|
||||
"crossFadeOffsetRate",
|
||||
"crossFadeEndRate",
|
||||
"serverInputAudioGain",
|
||||
"serverOutputAudioGain",
|
||||
]
|
||||
)
|
||||
strData: list[str] = field(default_factory=lambda: [])
|
||||
@ -108,120 +82,6 @@ class VoiceChanger:
|
||||
|
||||
# emitTo = None
|
||||
|
||||
# def audio_callback(self, indata: np.ndarray, outdata: np.ndarray, frames, times, status):
|
||||
# try:
|
||||
# indata = indata * self.settings.serverInputAudioGain
|
||||
# with Timer("all_inference_time") as t:
|
||||
# unpackedData = librosa.to_mono(indata.T) * 32768.0
|
||||
# out_wav, times = self.on_request(unpackedData)
|
||||
# outputChunnels = outdata.shape[1]
|
||||
# outdata[:] = np.repeat(out_wav, outputChunnels).reshape(-1, outputChunnels) / 32768.0
|
||||
# outdata[:] = outdata * self.settings.serverOutputAudioGain
|
||||
# all_inference_time = t.secs
|
||||
# performance = [all_inference_time] + times
|
||||
# if self.emitTo is not None:
|
||||
# self.emitTo(performance)
|
||||
# self.settings.performance = [round(x * 1000) for x in performance]
|
||||
# except Exception as e:
|
||||
# print("[Voice Changer] ex:", e)
|
||||
|
||||
# def getServerAudioDevice(self, audioDeviceList: list[ServerAudioDevice], index: int):
|
||||
# serverAudioDevice = [x for x in audioDeviceList if x.index == index]
|
||||
# if len(serverAudioDevice) > 0:
|
||||
# return serverAudioDevice[0]
|
||||
# else:
|
||||
# return None
|
||||
|
||||
# def serverLocal(self, _vc):
|
||||
# vc: VoiceChanger = _vc
|
||||
|
||||
# currentInputDeviceId = -1
|
||||
# currentModelSamplingRate = -1
|
||||
# currentOutputDeviceId = -1
|
||||
# currentInputChunkNum = -1
|
||||
# while True:
|
||||
# if vc.settings.serverAudioStated == 0 or vc.settings.serverInputDeviceId == -1 or vc.voiceChanger is None:
|
||||
# vc.settings.inputSampleRate = 48000
|
||||
# time.sleep(2)
|
||||
# else:
|
||||
# sd._terminate()
|
||||
# sd._initialize()
|
||||
|
||||
# sd.default.device[0] = vc.settings.serverInputDeviceId
|
||||
# currentInputDeviceId = vc.settings.serverInputDeviceId
|
||||
# sd.default.device[1] = vc.settings.serverOutputDeviceId
|
||||
# currentOutputDeviceId = vc.settings.serverOutputDeviceId
|
||||
|
||||
# currentInputChannelNum = vc.settings.serverAudioInputDevices
|
||||
|
||||
# serverInputAudioDevice = self.getServerAudioDevice(vc.settings.serverAudioInputDevices, currentInputDeviceId)
|
||||
# serverOutputAudioDevice = self.getServerAudioDevice(vc.settings.serverAudioOutputDevices, currentOutputDeviceId)
|
||||
# print(serverInputAudioDevice, serverOutputAudioDevice)
|
||||
# if serverInputAudioDevice is None or serverOutputAudioDevice is None:
|
||||
# time.sleep(2)
|
||||
# print("serverInputAudioDevice or serverOutputAudioDevice is None")
|
||||
# continue
|
||||
|
||||
# currentInputChannelNum = serverInputAudioDevice.maxInputChannels
|
||||
# currentOutputChannelNum = serverOutputAudioDevice.maxOutputChannels
|
||||
|
||||
# currentInputChunkNum = vc.settings.serverReadChunkSize
|
||||
# block_frame = currentInputChunkNum * 128
|
||||
|
||||
# # sample rate precheck(alsa cannot use 40000?)
|
||||
# try:
|
||||
# currentModelSamplingRate = self.voiceChanger.get_processing_sampling_rate()
|
||||
# except Exception as e:
|
||||
# print("[Voice Changer] ex: get_processing_sampling_rate", e)
|
||||
# continue
|
||||
# try:
|
||||
# with sd.Stream(
|
||||
# callback=self.audio_callback,
|
||||
# blocksize=block_frame,
|
||||
# samplerate=currentModelSamplingRate,
|
||||
# dtype="float32",
|
||||
# channels=[currentInputChannelNum, currentOutputChannelNum],
|
||||
# ):
|
||||
# pass
|
||||
# vc.settings.serverInputAudioSampleRate = currentModelSamplingRate
|
||||
# vc.settings.inputSampleRate = currentModelSamplingRate
|
||||
# print(f"[Voice Changer] sample rate {vc.settings.serverInputAudioSampleRate}")
|
||||
# except Exception as e:
|
||||
# print(
|
||||
# "[Voice Changer] ex: fallback to device default samplerate",
|
||||
# e,
|
||||
# )
|
||||
# vc.settings.serverInputAudioSampleRate = serverInputAudioDevice.default_samplerate
|
||||
# vc.settings.inputSampleRate = vc.settings.serverInputAudioSampleRate
|
||||
|
||||
# # main loop
|
||||
# try:
|
||||
# with sd.Stream(
|
||||
# callback=self.audio_callback,
|
||||
# blocksize=block_frame,
|
||||
# samplerate=vc.settings.serverInputAudioSampleRate,
|
||||
# dtype="float32",
|
||||
# channels=[currentInputChannelNum, currentOutputChannelNum],
|
||||
# ):
|
||||
# while vc.settings.serverAudioStated == 1 and currentInputDeviceId == vc.settings.serverInputDeviceId and currentOutputDeviceId == vc.settings.serverOutputDeviceId and currentModelSamplingRate == self.voiceChanger.get_processing_sampling_rate() and currentInputChunkNum == vc.settings.serverReadChunkSize:
|
||||
# time.sleep(2)
|
||||
# print(
|
||||
# "[Voice Changer] server audio",
|
||||
# self.settings.performance,
|
||||
# )
|
||||
# print(
|
||||
# "[Voice Changer] info:",
|
||||
# vc.settings.serverAudioStated,
|
||||
# currentInputDeviceId,
|
||||
# currentOutputDeviceId,
|
||||
# vc.settings.serverInputAudioSampleRate,
|
||||
# currentInputChunkNum,
|
||||
# )
|
||||
|
||||
# except Exception as e:
|
||||
# print("[Voice Changer] ex:", e)
|
||||
# time.sleep(2)
|
||||
|
||||
def __init__(self, params: VoiceChangerParams, slotIndex: int):
|
||||
# 初期化
|
||||
self.settings = VoiceChangerSettings()
|
||||
@ -238,9 +98,9 @@ class VoiceChanger:
|
||||
self.ioRecorder: IORecorder | None = None
|
||||
self.sola_buffer: AudioInOut | None = None
|
||||
|
||||
audioinput, audiooutput = list_audio_device()
|
||||
self.settings.serverAudioInputDevices = audioinput
|
||||
self.settings.serverAudioOutputDevices = audiooutput
|
||||
# audioinput, audiooutput = list_audio_device()
|
||||
# self.settings.serverAudioInputDevices = audioinput
|
||||
# self.settings.serverAudioOutputDevices = audiooutput
|
||||
|
||||
self.slotIndex = slotIndex
|
||||
self.slotInfo = loadSlotInfo(params.model_dir, self.slotIndex)
|
||||
|
||||
@ -3,6 +3,7 @@ import threading
|
||||
from data.ModelSample import ModelSamples
|
||||
from data.ModelSlot import ModelSlots, loadSlotInfo
|
||||
from utils.downloader.SampleDownloader import downloadSample, getSampleInfos
|
||||
from voice_changer.Local.AudioDeviceList import ServerAudioDevice, list_audio_device
|
||||
from voice_changer.Local.ServerDevice import ServerDevice
|
||||
from voice_changer.RVC.ModelSlotGenerator import setSlotAsRVC
|
||||
|
||||
@ -26,6 +27,7 @@ class GPUInfo:
|
||||
@dataclass()
|
||||
class VoiceChangerManagerSettings:
|
||||
slotIndex: int
|
||||
|
||||
intData: list[str] = field(default_factory=lambda: ["slotIndex"])
|
||||
|
||||
|
||||
@ -43,6 +45,14 @@ class VoiceChangerManager(object):
|
||||
self.sampleModels: list[ModelSamples] = getSampleInfos(self.params.sample_mode)
|
||||
self.gpus: list[GPUInfo] = self._get_gpuInfos()
|
||||
|
||||
audioinput, audiooutput = list_audio_device()
|
||||
self.serverAudioInputDevices: list[ServerAudioDevice] = audioinput
|
||||
self.serverAudioOutputDevices: list[ServerAudioDevice] = audiooutput
|
||||
|
||||
# ServerDevice
|
||||
thread = threading.Thread(target=self.serverDevice.serverLocal, args=(self,))
|
||||
thread.start()
|
||||
|
||||
def _get_gpuInfos(self):
|
||||
devCount = torch.cuda.device_count()
|
||||
gpus = []
|
||||
@ -63,8 +73,6 @@ class VoiceChangerManager(object):
|
||||
print(f"VoiceChanger Initialized (GPU_NUM:{gpu_num}, mps_enabled:{mps_enabled})")
|
||||
|
||||
cls._instance.voiceChanger = VoiceChanger(params, cls._instance.settings.slotIndex)
|
||||
thread = threading.Thread(target=cls._instance.serverDevice.serverLocal, args=(cls._instance.voiceChanger,))
|
||||
thread.start()
|
||||
cls._instance.voiceChanger.prepareModel()
|
||||
return cls._instance
|
||||
|
||||
@ -94,6 +102,8 @@ class VoiceChangerManager(object):
|
||||
data["slotInfos"] = slotInfos
|
||||
data["gpus"] = self.gpus
|
||||
data["sampleModels"] = self.sampleModels
|
||||
data["serverAudioInputDevices"] = self.serverAudioInputDevices
|
||||
data["serverAudioOutputDevices"] = self.serverAudioOutputDevices
|
||||
|
||||
data["status"] = "OK"
|
||||
if hasattr(self, "voiceChanger"):
|
||||
|
||||
Loading…
Reference in New Issue
Block a user