1berry/wokada-voice-changer
1
0
Fork 0
Code Issues Packages Projects Releases Wiki Activity

bugfix:

Browse Source 
diffusion svc server mode silence front value
This commit is contained in:
w-okada 2023-07-18 08:29:17 +09:00
parent b559582dc4
commit 0957fbc923
4 changed files with 6 additions and 6 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

5
server/voice_changer/DiffusionSVC/DiffusionSVC.py
View File

@ -104,7 +104,7 @@ class DiffusionSVC(VoiceChangerModel):
convertSize = convertSize + (128 - (convertSize % 128)) convertSize = convertSize + (128 - (convertSize % 128))
# バッファがたまっていない場合はzeroで補う # バッファがたまっていない場合はzeroで補う
generateFeatureLength = int(((convertSize / self.inputSampleRate) * self.slotInfo.samplingRate) / 512) + 1 generateFeatureLength = int(((convertSize / self.inputSampleRate) * self.slotInfo.samplingRate) / 512) + 1
if self.audio_buffer.shape[0] < convertSize: if self.audio_buffer.shape[0] < convertSize:
self.audio_buffer = np.concatenate([np.zeros([convertSize]), self.audio_buffer]) self.audio_buffer = np.concatenate([np.zeros([convertSize]), self.audio_buffer])
self.pitchf_buffer = np.concatenate([np.zeros(generateFeatureLength), self.pitchf_buffer]) self.pitchf_buffer = np.concatenate([np.zeros(generateFeatureLength), self.pitchf_buffer])
@ -151,7 +151,7 @@ class DiffusionSVC(VoiceChangerModel):
speedUp = self.settings.speedUp speedUp = self.settings.speedUp
embOutputLayer = 12 embOutputLayer = 12
useFinalProj = False useFinalProj = False
silenceFrontSec = self.settings.extraConvertSize / self.slotInfo.samplingRate if self.settings.silenceFront else 0. # extaraConvertSize(既にモデルのサンプリングレートにリサンプリング済み)の秒数。モデルのサンプリングレートで処理(★１)。 silenceFrontSec = self.settings.extraConvertSize / self.inputSampleRate if self.settings.silenceFront else 0. # extaraConvertSize(既にモデルのサンプリングレートにリサンプリング済み)の秒数。モデルのサンプリングレートで処理(★１)。
try: try:
audio_out, self.pitchf_buffer, self.feature_buffer = self.pipeline.exec( audio_out, self.pitchf_buffer, self.feature_buffer = self.pipeline.exec(
@ -169,7 +169,6 @@ class DiffusionSVC(VoiceChangerModel):
protect protect
) )
result = audio_out.detach().cpu().numpy() result = audio_out.detach().cpu().numpy()
return result return result
except DeviceCannotSupportHalfPrecisionException as e: # NOQA except DeviceCannotSupportHalfPrecisionException as e: # NOQA
print("[Device Manager] Device cannot support half precision. Fallback to float....") print("[Device Manager] Device cannot support half precision. Fallback to float....")

1
server/voice_changer/DiffusionSVC/inferencer/DiffusionSVCInferencer.py
View File

@ -125,7 +125,6 @@ class DiffusionSVCInferencer(Inferencer):
with Timer("pre-process") as t: # NOQA with Timer("pre-process") as t: # NOQA
start_frame = int(silence_front * self.vocoder.vocoder_sample_rate / self.vocoder.vocoder_hop_size) start_frame = int(silence_front * self.vocoder.vocoder_sample_rate / self.vocoder.vocoder_hop_size)
out_wav = self.mel2wav(out_mel, pitch, start_frame=start_frame) out_wav = self.mel2wav(out_mel, pitch, start_frame=start_frame)
out_wav *= mask out_wav *= mask
# print("[ ----Timer::3: ]", t.secs, start_frame, out_mel.shape) # print("[ ----Timer::3: ]", t.secs, start_frame, out_mel.shape)

4
server/voice_changer/Local/ServerDevice.py
View File

@ -271,7 +271,9 @@ class ServerDevice:
# Blockサイズを計算 # Blockサイズを計算
currentInputChunkNum = self.settings.serverReadChunkSize currentInputChunkNum = self.settings.serverReadChunkSize
block_frame = currentInputChunkNum * 128 # block_frame = currentInputChunkNum * 128
block_frame = int(currentInputChunkNum * 128 * (self.settings.serverInputAudioSampleRate / 48000))
sd.default.blocksize = block_frame sd.default.blocksize = block_frame
# main loop # main loop

2
server/voice_changer/VoiceChangerV2.py
View File

@ -260,7 +260,7 @@ class VoiceChangerV2(VoiceChangerIF):
print_convert_processing(f" Output data size of {result.shape[0]}/{processing_sampling_rate}hz {result .shape[0]}/{self.settings.outputSampleRate}hz") print_convert_processing(f" Output data size of {result.shape[0]}/{processing_sampling_rate}hz {result .shape[0]}/{self.settings.outputSampleRate}hz")
if receivedData.shape[0] != result .shape[0]: if receivedData.shape[0] != result.shape[0]:
outputData = pad_array(result, receivedData.shape[0]) outputData = pad_array(result, receivedData.shape[0])
pass pass
else: else: