azalea/VITS-fast-fine-tuning
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

upload files

Browse Source
This commit is contained in:
Plachta
2023-02-14 20:08:44 +08:00
parent eb546d3607
commit 7db09abc2e
1 changed files with 13 additions and 3 deletions
Download Patch File Download Diff File Expand all files Collapse all files
VC_inference.py
+13 -3
View File
@@ -23,7 +23,7 @@ def get_text(text, hps):
return text_norm return text_norm
def create_vc_fn(model, hps, speaker_ids): def create_vc_fn(model, hps, speaker_ids):
def vc_fn(original_speaker, target_speaker, record_audio, upload_audio): def vc_fn(original_speaker, target_speaker, record_audio, upload_audio, denoise):
input_audio = record_audio if record_audio is not None else upload_audio input_audio = record_audio if record_audio is not None else upload_audio
if input_audio is None: if input_audio is None:
return "You need to record or upload an audio", None return "You need to record or upload an audio", None
@@ -37,8 +37,17 @@ def create_vc_fn(model, hps, speaker_ids):
if sampling_rate != hps.data.sampling_rate: if sampling_rate != hps.data.sampling_rate:
audio = librosa.resample(audio, orig_sr=sampling_rate, target_sr=hps.data.sampling_rate) audio = librosa.resample(audio, orig_sr=sampling_rate, target_sr=hps.data.sampling_rate)
with no_grad(): with no_grad():
y = torch.FloatTensor(audio).to(device) y = torch.FloatTensor(audio)
y = y.unsqueeze(0) y = y.unsqueeze(0)
y = y / max(-y.min(), y.max()) / 0.99
if denoise:
torchaudio.save("infer.wav", y.cpu(), 22050, channels_first=True)
os.system(f"demucs --two-stems=vocals infer.wav")
y, sr = torchaudio.load(f"./separated/htdemucs/infer/vocals.wav", frame_offset=0, num_frames=-1, normalize=True, channels_first=True)
y = y.mean(dim=0).unsqueeze(0)
if sr != 22050:
y = torchaudio.transforms.Resample(orig_freq=sr, new_freq=22050)(y)
y = y.to(device)
spec = spectrogram_torch(y, hps.data.filter_length, spec = spectrogram_torch(y, hps.data.filter_length,
hps.data.sampling_rate, hps.data.hop_length, hps.data.win_length, hps.data.sampling_rate, hps.data.hop_length, hps.data.win_length,
center=False).to(device) center=False).to(device)
@@ -82,10 +91,11 @@ if __name__ == "__main__":
upload_audio = gr.Audio(label="or upload audio here", source="upload") upload_audio = gr.Audio(label="or upload audio here", source="upload")
source_speaker = gr.Dropdown(choices=speakers, value="User", label="source speaker") source_speaker = gr.Dropdown(choices=speakers, value="User", label="source speaker")
target_speaker = gr.Dropdown(choices=speakers, value=speakers[0], label="target speaker") target_speaker = gr.Dropdown(choices=speakers, value=speakers[0], label="target speaker")
denoise_checkbox = gr.Checkbox(label="denoise using demucs", value=True)
with gr.Column(): with gr.Column():
message_box = gr.Textbox(label="Message") message_box = gr.Textbox(label="Message")
converted_audio = gr.Audio(label='converted audio') converted_audio = gr.Audio(label='converted audio')
btn = gr.Button("Convert!") btn = gr.Button("Convert!")
btn.click(vc_fn, inputs=[source_speaker, target_speaker, record_audio, upload_audio], btn.click(vc_fn, inputs=[source_speaker, target_speaker, record_audio, upload_audio, denoise_checkbox],
outputs=[message_box, converted_audio]) outputs=[message_box, converted_audio])
app.launch(share=args.share) app.launch(share=args.share)