Update app.py

app.py

@@ -3,6 +3,7 @@ import os
 import re
 
 import torch
+import torchaudio
 
 import gradio as gr
 import spaces
@@ -21,6 +22,8 @@ lang = "no"
 
 logo_path = os.path.join(os.path.dirname(__file__), "Logo_2.png")
 
+max_audio_length= 1 * 60
+
 share = (os.environ.get("SHARE", "False")[0].lower() in "ty1") or None
 auth_token = os.environ.get("AUTH_TOKEN") or True
 device = torch.device("cuda:0" if torch.cuda.is_available() else "cpu")
@@ -52,11 +55,25 @@ def format_output(text):
     return text
 
 def transcribe(file, return_timestamps=False):
+    waveform, sample_rate = torchaudio.load(file)
+    audio_duration = waveform.size(1) / sample_rate  
+
+    if audio_duration > MAX_AUDIO_LENGTH:
+        # Trim the waveform to the first 30 minutes
+        waveform = waveform[:, :int(MAX_AUDIO_LENGTH * sample_rate)]
+        truncated_file = "truncated_audio.wav"
+        torchaudio.save(truncated_file, waveform, sample_rate)
+        file_to_transcribe = truncated_file
+        truncated = True
+    else:
+        file_to_transcribe = file
+        truncated = False
+        
     if not return_timestamps:
-        text = pipe(file)["text"]
+        text = pipe(file_to_transcribe)["text"]
         formatted_text = format_output(text)
     else:
-        chunks = pipe(file, return_timestamps=True)["chunks"]
+        chunks = pipe(file_to_transcribe, return_timestamps=True)["chunks"]
         text = []
         for chunk in chunks:
             start_time = time.strftime('%H:%M:%S', time.gmtime(chunk["timestamp"][0])) if chunk["timestamp"][0] is not None else "??:??:??"
@@ -64,7 +81,19 @@ def transcribe(file, return_timestamps=False):
             line = f"[{start_time} -> {end_time}] {chunk['text']}"
             text.append(line)
         formatted_text = "\n".join(text)
+
+    if truncated:
+        disclaimer = (
+            "\n\nDette er en demo. Det er ikke tillatt å bruke denne teksten i profesjonell sammenheng. "
+            "Vi anbefaler at hvis du trenger å transkribere lengre opptak, så kjører du enten modellen lokalt "
+            "eller sjekker denne siden for å se hvem som leverer løsninger basert på NB-Whisper: "
+            "https://github.com/NbAiLab/nostram/blob/main/leverandorer.md"
+        )
+        formatted_text += f"<br><br><i>{disclaimer}</i>"
+    
     formatted_text += "<br><br><i>Transkribert med NB-Whisper demo</i>"
+
+    
     return formatted_text
 
 def _return_yt_html_embed(yt_url):
@@ -101,13 +130,6 @@ demo = gr.Blocks()
 
 with demo:
     with gr.Row():
-        #with gr.Column(scale=1, min_width=120):
-            #gr.Markdown(
-                #f"""
-                #gr.Markdown("!local img[]({logo_path})")
-
-                #"""
-            #)
         gr.HTML("<img src='file/Logo_2.png'>") 
         with gr.Column(scale=8):
             # Use Markdown for title and description