Update app.py

app.py

@@ -18,33 +18,33 @@ def get_wavlm():
 
 device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
 
-print("Loading FreeVC...")
-hps = utils.get_hparams_from_file("configs/freevc.json")
+print("Loading CloneVoiceAI...")
+hps = utils.get_hparams_from_file("configs/CloneVoiceAI.json")
 freevc = SynthesizerTrn(
     hps.data.filter_length // 2 + 1,
     hps.train.segment_size // hps.data.hop_length,
     **hps.model).to(device)
 _ = freevc.eval()
-_ = utils.load_checkpoint("checkpoints/freevc.pth", freevc, None)
+_ = utils.load_checkpoint("checkpoints/CloneVoiceAI.pth", freevc, None)
 smodel = SpeakerEncoder('speaker_encoder/ckpt/pretrained_bak_5805000.pt')
 
-print("Loading FreeVC(24k)...")
-hps = utils.get_hparams_from_file("configs/freevc-24.json")
+print("Loading CloneVoiceAI(24k)...")
+hps = utils.get_hparams_from_file("configs/CloneVoiceAI-24.json")
 freevc_24 = SynthesizerTrn(
     hps.data.filter_length // 2 + 1,
     hps.train.segment_size // hps.data.hop_length,
     **hps.model).to(device)
 _ = freevc_24.eval()
-_ = utils.load_checkpoint("checkpoints/freevc-24.pth", freevc_24, None)
+_ = utils.load_checkpoint("checkpoints/CloneVoiceAI-24.pth", freevc_24, None)
 
-print("Loading FreeVC-s...")
-hps = utils.get_hparams_from_file("configs/freevc-s.json")
+print("Loading CloneVoiceAI-s...")
+hps = utils.get_hparams_from_file("configs/CloneVoiceAI-s.json")
 freevc_s = SynthesizerTrn(
     hps.data.filter_length // 2 + 1,
     hps.train.segment_size // hps.data.hop_length,
     **hps.model).to(device)
 _ = freevc_s.eval()
-_ = utils.load_checkpoint("checkpoints/freevc-s.pth", freevc_s, None)
+_ = utils.load_checkpoint("checkpoints/CloneVoiceAI-s.pth", freevc_s, None)
 
 print("Loading WavLM for content...")
 cmodel = WavLMModel.from_pretrained("microsoft/wavlm-large").to(device)
@@ -54,7 +54,7 @@ def convert(model, src, tgt):
         # tgt
         wav_tgt, _ = librosa.load(tgt, sr=hps.data.sampling_rate)
         wav_tgt, _ = librosa.effects.trim(wav_tgt, top_db=20)
-        if model == "FreeVC" or model == "FreeVC (24kHz)":
+        if model == "CloneVoiceAI" or model == "CloneVoiceAI (24kHz)":
             g_tgt = smodel.embed_utterance(wav_tgt)
             g_tgt = torch.from_numpy(g_tgt).unsqueeze(0).to(device)
         else:
@@ -74,30 +74,30 @@ def convert(model, src, tgt):
         wav_src = torch.from_numpy(wav_src).unsqueeze(0).to(device)
         c = cmodel(wav_src).last_hidden_state.transpose(1, 2).to(device)
         # infer
-        if model == "FreeVC":
+        if model == "CloneVoiceAI":
             audio = freevc.infer(c, g=g_tgt)
-        elif model == "FreeVC-s":
+        elif model == "CloneVoiceAI-s":
             audio = freevc_s.infer(c, mel=mel_tgt)
         else:
             audio = freevc_24.infer(c, g=g_tgt)
         audio = audio[0][0].data.cpu().float().numpy()
-        if model == "FreeVC" or model == "FreeVC-s":
+        if model == "CloneVoiceAI" or model == "CloneVoiceAI-s":
             write("out.wav", hps.data.sampling_rate, audio)
         else:
             write("out.wav", 24000, audio)
     out = "out.wav"
     return out
     
-model = gr.Dropdown(choices=["FreeVC", "FreeVC-s", "FreeVC (24kHz)"], value="FreeVC",type="value", label="Model") 
+model = gr.Dropdown(choices=["CloneVoiceAI", "CloneVoiceAI-s", "CloneVoiceAI (24kHz)"], value="CloneVoiceAI",type="value", label="Model") 
 audio1 = gr.Audio(label="Source Audio", type='filepath')
 audio2 = gr.Audio(label="Reference Audio", type='filepath')
 inputs = [model, audio1, audio2]
 outputs =  gr.Audio(label="Output Audio", type='filepath')
 
-title = "FreeVC"
-description = "Gradio Demo for FreeVC: Towards High-Quality Text-Free One-Shot Voice Conversion. To use it, simply upload your audio, or click the example to load. Read more at the links below. Note: It seems that the WavLM checkpoint in HuggingFace is a little different from the one used to train FreeVC, which may degrade the performance a bit. In addition, speaker similarity can be largely affected if there are too much silence in the reference audio, so please <strong>trim</strong> it before submitting."
-article = "<p style='text-align: center'><a href='https://arxiv.org/abs/2210.15418' target='_blank'>Paper</a> | <a href='https://github.com/OlaWod/FreeVC' target='_blank'>Github Repo</a></p>"
+title = "CloneVoiceAI"
+description = "Gradio Demo for CloneVoiceAI: Towards High-Quality Text-Free One-Shot Voice Conversion. To use it, simply upload your audio, or click the example to load. Read more at the links below. Note: It seems that the WavLM checkpoint in HuggingFace is a little different from the one used to train FreeVC, which may degrade the performance a bit. In addition, speaker similarity can be largely affected if there are too much silence in the reference audio, so please <strong>trim</strong> it before submitting."
+#article = "<p style='text-align: center'><a href='https://arxiv.org/abs/2210.15418' target='_blank'>Paper</a> | <a href='https://github.com/OlaWod/FreeVC' target='_blank'>Github Repo</a></p>"
 
-examples=[["FreeVC", 'p225_001.wav', 'p226_002.wav'], ["FreeVC-s", 'p226_002.wav', 'p225_001.wav'], ["FreeVC (24kHz)", 'p225_001.wav', 'p226_002.wav']]
+examples=[["CloneVoiceAI", 'p225_001.wav', 'p226_002.wav'], ["CloneVoiceAI-s", 'p226_002.wav', 'p225_001.wav'], ["CloneVoiceAI (24kHz)", 'p225_001.wav', 'p226_002.wav']]
 
 gr.Interface(convert, inputs, outputs, title=title, description=description, article=article, examples=examples).launch()