Update app.py

app.py

@@ -1,10 +1,84 @@
 import gradio as gr
-from huggingface_hub import InferenceClient
+from huggingface_hub import hf_hub_download
+
 
 """
 For more information on `huggingface_hub` Inference API support, please check the docs: https://huggingface.co/docs/huggingface_hub/v0.22.2/en/guides/inference
 """
-client = InferenceClient("HuggingFaceH4/zephyr-7b-beta")
+
+
+import os
+import pickle
+import numpy as np
+import torch
+import torch.nn.functional as F
+from collections import OrderedDict
+from AutoPST.onmt_modules.misc import sequence_mask
+from AutoPST.model_autopst import Generator_2 as Predictor
+from AutoPST.hparams_autopst import hparams
+
+device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
+
+P = Predictor(hparams).eval().to(device)
+
+checkpoint = torch.load(hf_hub_download(repo_id="jonathanjordan21/AutoPST", filename='580000-P.ckpt'), map_location=lambda storage, loc: storage)  
+P.load_state_dict(checkpoint['model'], strict=True)
+print('Loaded predictor .....................................................')
+
+dict_test = pickle.load(open('./AutoPST/assets/test_vctk.meta', 'rb'))
+
+spect_vc = OrderedDict()
+
+uttrs = [('p231', 'p270', '001'),
+         ('p270', 'p231', '001'),
+         ('p231', 'p245', '003001'),
+         ('p245', 'p231', '003001'),
+         ('p239', 'p270', '024002'),
+         ('p270', 'p239', '024002')]
+
+
+for uttr in uttrs:
+        
+    cep_real, spk_emb = dict_test[uttr[0]][uttr[2]]
+    cep_real_A = torch.from_numpy(cep_real).unsqueeze(0).to(device)
+    len_real_A = torch.tensor(cep_real_A.size(1)).unsqueeze(0).to(device)
+    real_mask_A = sequence_mask(len_real_A, cep_real_A.size(1)).float()
+    
+    _, spk_emb = dict_test[uttr[1]][uttr[2]]
+    spk_emb_B = torch.from_numpy(spk_emb).unsqueeze(0).to(device)
+    
+    with torch.no_grad():
+        spect_output, len_spect = P.infer_onmt(cep_real_A.transpose(2,1)[:,:14,:],
+                                               real_mask_A,
+                                               len_real_A,
+                                               spk_emb_B)
+    
+    uttr_tgt = spect_output[:len_spect[0],0,:].cpu().numpy()
+        
+    spect_vc[f'{uttr[0]}_{uttr[1]}_{uttr[2]}'] = uttr_tgt
+
+# spectrogram to waveform
+# Feel free to use other vocoders
+# This cell requires some preparation to work, please see the corresponding part in AutoVC
+import torch
+import librosa
+import pickle
+import os
+from AutoPST.synthesis import build_model
+from AutoPST.synthesis import wavegen
+
+model = build_model().to(device)
+checkpoint = torch.load(hf_hub_download(repo_id="jonathanjordan21/AutoPST", filename="checkpoint_step001000000_ema.pth"), map_location=torch.device('cpu'))
+model.load_state_dict(checkpoint["state_dict"])
+
+# for name, sp in spect_vc.items():
+
+#     print(name)
+#     waveform = wavegen(model, c=sp)   
+
+#     librosa.output.write_wav('./assets/'+name+'.wav', waveform, sr=16000)
+
+
 
 
 def respond(