Update app.py

app.py

@@ -32,6 +32,7 @@ from sklearn.metrics.pairwise import cosine_similarity
 import pandas as pd
 import numpy as np
 import re
+import spaces
 from transformers import AutoTokenizer, AutoModelForSeq2SeqLM
 import gradio as gr
 from fpdf import FPDF
@@ -45,6 +46,7 @@ PLACEHOLDER = """
    <p style="font-size: 18px; margin-bottom: 2px; opacity: 0.65;">En webapp for transkribering av lydfiler til norsk skrift. Språkmodell: NbAiLab/nb-whisper-large, Ekstra: oppsummering, pdf-download</p>
 </div>
 """
+
 # Suppress warnings
 warnings.filterwarnings("ignore")
 
@@ -114,10 +116,7 @@ def transcribe_audio(audio_file, chunk_length_s=30):
 
         
         # Tokenize the input batch with the processor
-        inputs = processor(chunk_waveform.squeeze(0).numpy(), sampling_rate=sample_rate, return_tensors="pt", task="transcribe")
-
-        # Use the attention mask directly from the tokenizer output
-        attention_mask = inputs.attention_mask.to(device)
+        inputs = processor(chunk_waveform.squeeze(0).numpy(), sampling_rate=sample_rate, padding="max_length", return_tensors="pt", task="transcribe", device=device)
 
         # ASR model inference on the chunk
         with torch.no_grad():