Update app.py

app.py

@@ -1,4 +1,11 @@
-import os
+import networkx as nx
+from sklearn.feature_extraction.text import TfidfVectorizer
+from sklearn.metrics.pairwise import cosine_similarity
+import nltk
+nltk.download('punkt')
+from nltk.tokenize import sent_tokenize
+
+
 import gradio as gr
 import warnings
 import torch
@@ -6,31 +13,26 @@ from transformers import WhisperTokenizer, WhisperForConditionalGeneration, Whis
 from pydub import AudioSegment
 import soundfile as sf
 import numpy as np
-import nltk
 from fpdf import FPDF
 import time
+import os
 
 warnings.filterwarnings("ignore")
-nltk.download('punkt')
 
-# Load environment variable
 HF_AUTH_TOKEN = os.getenv('HF_AUTH_TOKEN')
 
-# Load tokenizer and model
+
 tokenizer = WhisperTokenizer.from_pretrained("NbAiLabBeta/nb-whisper-medium")
 model = WhisperForConditionalGeneration.from_pretrained("NbAiLabBeta/nb-whisper-medium")
 processor = WhisperProcessor.from_pretrained("NbAiLabBeta/nb-whisper-medium")
 
-# Set up the device
+
 device = torch.device('cuda' if torch.cuda.is_available() else 'cpu')
-torch_dtype = torch.float32
 model.to(device)
 
-# Initialize pipeline
 asr = pipeline("automatic-speech-recognition", model=model, tokenizer=processor.tokenizer, feature_extractor=processor.feature_extractor, device=device, torch_dtype=torch.float32)
 
 def transcribe_audio(audio_file):
-    # Perform transcription
     with torch.no_grad():
         output = asr(audio_file, chunk_length_s=28, generate_kwargs={"num_beams": 5, "task": "transcribe", "language": "no"})
     return output["text"]
@@ -95,11 +97,24 @@ def transcribe_audio(audio_file, batch_size=4):
     
     return transcription.strip(), result
 
+# Graph-based summarization|TextRank
 def summarize_text(text):
-    inputs = summarization_tokenizer("summarize: " + text, return_tensors="pt", max_length=1024, truncation=True)
-    inputs = inputs.to(device)
-    summary_ids = summarization_model.generate(inputs.input_ids, num_beams=4, max_length=150, early_stopping=True)
-    summary = summarization_tokenizer.decode(summary_ids[0], skip_special_tokens=True)
+    sentences = sent_tokenize(text)
+    if len(sentences) == 0:
+        return ""
+    
+    tfidf_vectorizer = TfidfVectorizer()
+    tfidf_matrix = tfidf_vectorizer.fit_transform(sentences)
+    similarity_matrix = cosine_similarity(tfidf_matrix)
+
+    nx_graph = nx.from_numpy_array(similarity_matrix)
+    scores = nx.pagerank(nx_graph)
+
+    ranked_sentences = sorted(((scores[i], s) for i, s in enumerate(sentences)), reverse=True)
+
+    # Select top N sentences (e.g., 3 sentences for the summary)
+    top_n = 3
+    summary = " ".join([s for _, s in ranked_sentences[:top_n]])
     return summary
 
 # HTML syntax for imagery