PublicMirrors/voice-changer
1
0
Fork 0
mirror of https://github.com/w-okada/voice-changer.git synced 2025-02-09 03:37:51 +03:00
Code Issues Actions Packages Projects Releases Wiki Activity

cpu

Browse Source
This commit is contained in:
wataru 2023-01-05 18:56:02 +09:00
parent 07a14e6bcd
commit 8486076779
4 changed files with 25 additions and 11 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

2
client/demo/src/index.tsx
View File

@ -149,7 +149,7 @@ const App = () => {
<> <>
<div className="body-row split-3-1-1-1-4 left-padding-1 highlight"> <div className="body-row split-3-1-1-1-4 left-padding-1 highlight">
<div className="body-item-title">monitor:</div> <div className="body-item-title">monitor:</div>
<div className="body-item-text">vol(db):{volume.toFixed(4)}</div> <div className="body-item-text">vol(rms):{volume.toFixed(4)}</div>
<div className="body-item-text">buf(ms):{bufferingTime}</div> <div className="body-item-text">buf(ms):{bufferingTime}</div>
<div className="body-item-text">res(ms):{responseTime}</div> <div className="body-item-text">res(ms):{responseTime}</div>
<div className="body-item-text"></div> <div className="body-item-text"></div>

5
client/demo/src/options_microphone.tsx
View File

@ -133,12 +133,9 @@ export const useMicrophoneOptions = (audioContext?: AudioContext): MicrophoneOpt
}, [inputAudioDeviceInfo, audioInput]) }, [inputAudioDeviceInfo, audioInput])
const audioMediaInputRow = useMemo(() => { const audioMediaInputRow = useMemo(() => {
console.log("GEN:audioMediaInputRow1")
if (audioInput != "file") { if (audioInput != "file") {
console.log("GEN:audioMediaInputRow2")
return <></> return <></>
} }
console.log("GEN:audioMediaInputRow3")
const onFileLoadClicked = async () => { const onFileLoadClicked = async () => {
const url = await fileSelectorAsDataURL("") const url = await fileSelectorAsDataURL("")
@ -185,7 +182,7 @@ export const useMicrophoneOptions = (audioContext?: AudioContext): MicrophoneOpt
</div> </div>
) )
}, [audioInput, audioOutput]) }, [audioInput, audioOutput])
console.log("GEN:audioMediaInputRow3")
useEffect(() => { useEffect(() => {
if (!audioContext) { if (!audioContext) {
return return

2
client/lib/worklet/src/voice-changer-worklet-processor.ts
View File

@ -58,7 +58,7 @@ class VoiceChangerWorkletProcessor extends AudioWorkletProcessor {
} }
if (this.playBuffer.length === 0) { if (this.playBuffer.length === 0) {
console.log("[worklet] no play buffer") // console.log("[worklet] no play buffer")
return true return true
} }

27
server/voice_changer/VoiceChanger.py
View File

@ -112,8 +112,25 @@ class VoiceChanger():
x, x_lengths, spec, spec_lengths, y, y_lengths, sid_src = [ x, x_lengths, spec, spec_lengths, y, y_lengths, sid_src = [
x.cpu() for x in data] x.cpu() for x in data]
sid_tgt1 = torch.LongTensor([dstId]).cpu() sid_tgt1 = torch.LongTensor([dstId]).cpu()
audio1 = (self.net_g.cpu().voice_conversion(spec, spec_lengths, sid_src=sid_src, sid_tgt=sid_tgt1)[ audio1 = (self.net_g.cpu().voice_conversion(spec, spec_lengths, sid_src=sid_src, sid_tgt=sid_tgt1)[0][0, 0].data * self.hps.data.max_wav_value)
0][0, 0].data * self.hps.data.max_wav_value).cpu().float().numpy()
if self.prev_strength.device != torch.device('cpu'):
print(f"prev_strength move from {self.prev_strength.device} to cpu")
self.prev_strength = self.prev_strength.cpu()
if self.cur_strength.device != torch.device('cpu'):
print(f"cur_strength move from {self.cur_strength.device} to cpu")
self.cur_strength = self.cur_strength.cpu()
if hasattr(self, 'prev_audio1') == True and self.prev_audio1.device == torch.device('cpu'):
prev = self.prev_audio1[-1*unpackedData.shape[0]:]
cur = audio1[-2*unpackedData.shape[0]:-1*unpackedData.shape[0]]
result = prev * self.prev_strength + cur * self.cur_strength
else:
cur = audio1[-2*unpackedData.shape[0]:-1*unpackedData.shape[0]]
result = cur
self.prev_audio1 = audio1
result = result.cpu().float().numpy()
# elif self.mps_enabled == True: # MPS doesnt support aten::weight_norm_interface, and PYTORCH_ENABLE_MPS_FALLBACK=1 cause a big dely. # elif self.mps_enabled == True: # MPS doesnt support aten::weight_norm_interface, and PYTORCH_ENABLE_MPS_FALLBACK=1 cause a big dely.
# x, x_lengths, spec, spec_lengths, y, y_lengths, sid_src = [ # x, x_lengths, spec, spec_lengths, y, y_lengths, sid_src = [
# x.to("mps") for x in data] # x.to("mps") for x in data]
@ -137,15 +154,15 @@ class VoiceChanger():
if hasattr(self, 'prev_audio1') == True: if hasattr(self, 'prev_audio1') == True and self.prev_audio1.device == torch.device('cuda', gpu):
prev = self.prev_audio1[-1*unpackedData.shape[0]:] prev = self.prev_audio1[-1*unpackedData.shape[0]:]
cur = audio1[-2*unpackedData.shape[0]:-1*unpackedData.shape[0]] cur = audio1[-2*unpackedData.shape[0]:-1*unpackedData.shape[0]]
result = prev * self.prev_strength + cur * self.cur_strength result = prev * self.prev_strength + cur * self.cur_strength
# print("merging...", prev.shape, cur.shape) print("merging...", prev.shape, cur.shape)
else: else:
cur = audio1[-2*unpackedData.shape[0]:-1*unpackedData.shape[0]] cur = audio1[-2*unpackedData.shape[0]:-1*unpackedData.shape[0]]
result = cur result = cur
# print("no merging...", cur.shape) print("no merging...", cur.shape)
self.prev_audio1 = audio1 self.prev_audio1 = audio1
#print(result) #print(result)