final commit

g_01720000 → CNN_SGD_001_1885.pt

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:fa0a76a49573909b968708717138849d68d9627cdf03307d00c7bd49278dc573
-size 55824433
+oid sha256:6ac783b9b049185deaa258d2c2c8ba1b1311451ea9fa74e83919130b529f2b2a
+size 1748943

app.py

@@ -39,7 +39,9 @@ model = SpecClassifier(
     model_type=getattr(hps.model, "classifier_type", "CNN-with-time")
 )
 
-ckpt = './cnnwt_SGD_1959.pt'
+# ckpt = './cnnwt_SGD_1959.pt'
+# ckpt_tts = './grad_uncond_cnn_001.pt'
+ckpt = './CNN_SGD_001_1885.pt'
 ckpt_tts = './grad_uncond_cnn_001.pt'
 
 utils.load_checkpoint_no_logger(ckpt_tts, gradtts_uncond_model, None)
@@ -70,7 +72,7 @@ def generate_audio(text, quantity, speaker, emotion_1, emotion_2):
 
     y_enc, y_dec, attn = gradtts_uncond_model.classifier_guidance_decode_two_mixture(
                         x, x_lengths,
-                        n_timesteps=10,
+                        n_timesteps=100,
                         temperature=2.0,
                         stoc=args.stoc,
                         spk=sid,
@@ -83,9 +85,8 @@ def generate_audio(text, quantity, speaker, emotion_1, emotion_2):
                         classifier_type=model.model_type
                     )
     y_dec = y_dec.detach()
-    # y_dec = torch.nan_to_num(y_dec)
-    res = y_dec.squeeze().cpu().numpy()
-    x = torch.from_numpy(res).cuda().unsqueeze(0)
+    res = y_dec.squeeze().to(device).numpy()
+    x = torch.from_numpy(res).unsqueeze(0)
     y_g_hat = vocoder(x)
     audio = y_g_hat.squeeze()
     audio = audio * 32768.0
@@ -102,8 +103,8 @@ spekears = ['Madi', 'Marzhan', 'Akzhol']
 demo = gr.Interface(
     generate_audio,
     [
-        gr.Textbox(value='Сәлем', label="Text you want to synthesize"),
-        gr.Slider(0, 100, value=0, step=10, label="Count", info="Choose between 0 and 100"),
+        gr.Textbox(value='Батпақ соры шабындыққа және жыл бойғы жайылымға пайдаланылады.', label="Text you want to synthesize"),
+        gr.Slider(0, 100, value=50, step=10, label="Count", info="Choose between 0 and 100"),
         gr.Dropdown(spekears, value=spekears[1], label="Narrator", info="Select a narrator."
         ),
         gr.Dropdown(emotions, label="Emotion 1", info="Select first emotion"),