Update app.py

app.py

@@ -111,11 +111,13 @@ def transcribe_audio(audio_file, chunk_length_s=30):
         # Create attention mask
         attention_mask = torch.ones(inputs.input_features.shape[:2], dtype=torch.long, device=device)
 
-        # input_ids used to set attention_mask to zero for padding tokens
-        input_ids = inputs['input_ids']  
+        # -- does not output input_ids (i.e, processor)
+        # input_ids = inputs['input_ids']  
+        # attention_mask[input_ids == processor.tokenizer.pad_token_id] = 0
+        
 
         # Set the attention mask to zero for padding tokens
-        attention_mask[input_ids == processor.tokenizer.pad_token_id] = 0
+        attention_mask[inputs.input_features.squeeze(0) == processor.tokenizer.pad_token_id] = 0
 
 
         # ASR model inference on the chunk
@@ -267,8 +269,8 @@ def save_to_pdf(text, summary):
 iface = gr.Blocks()
 
 with iface:
-    gr.HTML('<img src="https://huggingface.co/spaces/camparchimedes/ola_s-audioshop/blob/main/banner_trans.png" alt="banner exists!" width="100%" height="auto"/>')
-    gr.Markdown("**Switch Work's JoJo-versjon som webapp for transkribering av lydfiler til norsk skrift. Språkmodell: NbAiLab/nb-whisper-large, Ekstra: oppsummering, pdf-download**")
+    gr.HTML('<img src="https://huggingface.co/spaces/camparchimedes/ola_s-audioshop/blob/main/pic09w9678yhit.png" alt="" width="100%" height="auto"/>')
+    gr.Markdown("**Switch Work webapp for transkribering av lydfiler til norsk skrift. Språkmodell: NbAiLab/nb-whisper-large, Ekstra: oppsummering, pdf-download**")
 
     with gr.Tabs():
         with gr.TabItem("Transcription"):