Update app.py

app.py

@@ -1,17 +1,30 @@
 import streamlit as st
 import whisper
+import torch
 from transformers import pipeline
 import spacy
 from summa import keywords
 import datetime
 import os
+from pydub import AudioSegment
+import concurrent.futures
 
 @st.cache_resource
 def load_models():
-    whisper_model = whisper.load_model("base")  # You can use 'small' or 'medium' for better results
+    device = "cuda" if torch.cuda.is_available() else "cpu"
+    whisper_model = whisper.load_model("small").to(device)  # Using 'small' for faster speed
     summarizer = pipeline("summarization", model="facebook/bart-large-cnn")
     nlp = spacy.load("en_core_web_sm")
-    return whisper_model, summarizer, nlp
+    return whisper_model, summarizer, nlp, device
+
+def split_audio(file_path, chunk_length_ms=60000):  # 60 seconds per chunk
+    audio = AudioSegment.from_file(file_path)
+    chunks = [audio[i : i + chunk_length_ms] for i in range(0, len(audio), chunk_length_ms)]
+    return chunks
+
+def transcribe_chunk(whisper_model, chunk_path, device):
+    options = {"fp16": False} if device == "cpu" else {"fp16": True}
+    return whisper_model.transcribe(chunk_path, **options)["text"]
 
 def extract_action_items(text, nlp):
     doc = nlp(text)
@@ -39,30 +52,33 @@ def extract_action_items(text, nlp):
 def main():
     st.title("🤖 Smart AI Meeting Assistant")
     
-    whisper_model, summarizer, nlp = load_models()
+    whisper_model, summarizer, nlp, device = load_models()
     
     audio_file = st.file_uploader("Upload meeting audio", type=["wav", "mp3", "m4a", "ogg", "flac"])
     
     if audio_file is not None:
         file_path = f"uploaded_audio_{datetime.datetime.now().timestamp()}.wav"
         
-        # Save uploaded file
         with open(file_path, "wb") as f:
             f.write(audio_file.getbuffer())
 
         st.subheader("Meeting Transcription")
         with st.spinner("Transcribing audio..."):
-            # Load and process audio
-            audio = whisper.load_audio(file_path)  # Converts to 16kHz mono
-            audio = whisper.pad_or_trim(audio)  # Ensures proper input size
-            mel = whisper.log_mel_spectrogram(audio).to(whisper_model.device)
+            chunks = split_audio(file_path)
+            chunk_paths = []
+            
+            for i, chunk in enumerate(chunks):
+                chunk_path = f"chunk_{i}.wav"
+                chunk.export(chunk_path, format="wav")
+                chunk_paths.append(chunk_path)
+            
+            with concurrent.futures.ThreadPoolExecutor() as executor:
+                transcripts = list(executor.map(lambda cp: transcribe_chunk(whisper_model, cp, device), chunk_paths))
             
-            # Transcribe
-            result = whisper_model.transcribe(file_path)
-            transcript = result["text"]
+            transcript = " ".join(transcripts)
         
         st.write(transcript)
-        os.remove(file_path)  # Cleanup
+        os.remove(file_path)
         
         st.subheader("Meeting Summary")
         with st.spinner("Generating summary..."):