Update app.py

app.py

@@ -3,7 +3,7 @@ from model import Wav2VecModel
 from dataset import S2IDataset, collate_fn
 import requests
 requests.packages.urllib3.disable_warnings()
-
+import gradio as gr
 import torch
 import torch.nn as nn
 import torchaudio
@@ -103,72 +103,12 @@ class LightningModel(pl.LightningModule):
         return predicted_class
 
 if __name__ == "__main__":
-
-    dataset = S2IDataset(
-        csv_path="./speech-to-intent/train.csv",
-        wav_dir_path="/home/development/pavan/Telesoft/speech-to-intent-dataset/baselines/speech-to-intent"
-    )
-
-    test_dataset = S2IDataset(
-        csv_path="./speech-to-intent/test.csv",
-        wav_dir_path="/home/development/pavan/Telesoft/speech-to-intent-dataset/baselines/speech-to-intent"
-    )
-
-    train_len = int(len(dataset) * 0.90)
-    val_len =  len(dataset) - train_len
-    print(train_len, val_len)
-    train_dataset, val_dataset = torch.utils.data.random_split(dataset, [train_len, val_len], generator=torch.Generator().manual_seed(SEED))
-    print(len(test_dataset))
-
-    trainloader = torch.utils.data.DataLoader(
-            train_dataset, 
-            batch_size=4, 
-            shuffle=True, 
-            num_workers=4,
-            collate_fn = collate_fn,
-        )
-    
-    valloader = torch.utils.data.DataLoader(
-            val_dataset, 
-            batch_size=4, 
-            num_workers=4,
-            collate_fn = collate_fn,
-        )
-
-    testloader = torch.utils.data.DataLoader(
-            test_dataset, 
-            #batch_size=4, 
-            num_workers=4,
-            collate_fn = collate_fn,
-        )
-
     print(torch.cuda.mem_get_info())
 
     model = LightningModel()
 
     run_name = "wav2vec"
-    logger = WandbLogger(
-        name=run_name,
-        project='S2I-baseline'
-    )
-
-    model_checkpoint_callback = ModelCheckpoint(
-            dirpath='checkpoints',
-            monitor='val/acc', 
-            mode='max',
-            verbose=1,
-            filename=run_name + "-epoch={epoch}.ckpt")
 
-    trainer = Trainer(
-            fast_dev_run=False, 
-            gpus=1,
-            max_epochs=5, 
-            checkpoint_callback=True,
-            callbacks=[
-                model_checkpoint_callback,
-            ],
-            logger=logger,
-            )
     checkpoint_path = "./checkpoints/wav2vec-epoch=epoch=4.ckpt.ckpt"
     checkpoint = torch.load(checkpoint_path)
     model.load_state_dict(checkpoint['state_dict'])
@@ -187,6 +127,29 @@ if __name__ == "__main__":
     #with torch.no_grad():
      #   y_hat = model(wav_tensor)
 
+    def trabscribe(audio):
+        wav_tensor,_ = audio
+        wav_tensor = resmaple(wav_tensor)
+        #model = model.to('cuda') 
+        y_hat = model.predict(wav_tensor)
+        labels = {0:"branch_address : enquiry about bank branch location",
+        1:"activate_card : enquiry about activating card products",
+        2:"past_transactions : enquiry about past transactions in a specific time period",
+        3:"dispatch_status : enquiry about the dispatch status of card products",
+        4:"outstanding_balance : enquiry about outstanding balance on card products",
+        5:"card_issue : report about an issue with using card products",
+        6:"ifsc_code : enquiry about IFSC code of bank branch",
+        7:"generate_pin : enquiry about changing or generating a new pin for their card product",
+        8:"unauthorised_transaction : report about an unauthorised or fraudulent transaction",
+        9:"loan_query : enquiry about different kinds of loans",
+        10:"balance_enquiry : enquiry about bank account balance",
+        11:"change_limit : enquiry about changing the limit for card products",
+        12:"block : enquiry about blocking card or banking product",
+        13:"lost : report about losing a card product}
+        return labels[y_hat]
+
     print(y_hat)
+    get_intent = gr.Interface(fn = transcribe,
+                           gr.Audio(source="microphone"), outputs="text").launch()