Update app.py

app.py

@@ -4,53 +4,48 @@ from transformers import pipeline
 from pydub import AudioSegment
 import os
 
-# Set Streamlit page config
-st.set_page_config(page_title="Atma.ai - Session Summarizer", layout="centered")
+# Page config
+st.set_page_config(page_title="Atma.ai - Advanced Session Summarizer", layout="centered")
 
-st.title("🧠 Atma.ai – Mental Health Session Summarizer")
-st.markdown("Upload a therapy session audio file to get a transcript, summary, and emotional insights.")
+st.title("🧠 Atma.ai – Advanced Mental Health Session Summarizer")
+st.markdown("Upload a recorded therapy session to get a structured summary and emotional tone analysis. Now enhanced with dialogue-aware summarization!")
 
 # Upload audio
-uploaded_file = st.file_uploader("🎙️ Upload audio", type=["wav", "mp3", "m4a"])
+uploaded_file = st.file_uploader("🎙️ Upload audio file", type=["wav", "mp3", "m4a"])
 
 if uploaded_file:
     st.audio(uploaded_file)
 
-    # Save and convert audio
+    # Convert audio to required format
     audio_path = "temp_audio.wav"
     audio = AudioSegment.from_file(uploaded_file)
     audio = audio.set_channels(1).set_frame_rate(16000)
     audio.export(audio_path, format="wav")
 
     # Transcribe
-    st.info("Transcribing audio using Whisper...")
+    st.info("🔄 Transcribing with Whisper...")
     asr = pipeline("automatic-speech-recognition", model="openai/whisper-small")
     result = asr(audio_path, return_timestamps=True)
     transcript = result["text"]
 
     st.subheader("📝 Transcript")
-    st.write(transcript)
+    st.markdown(transcript)
 
-    # Summarize
-    st.info("Generating summary...")
-    summarizer = pipeline("summarization", model="facebook/bart-large-cnn")
-    summary = summarizer(transcript, max_length=250, min_length=50, do_sample=False)
+    # Dialogue-aware summarization using SAMSum-tuned model
+    st.info("📋 Summarizing conversation contextually...")
+    summarizer = pipeline("summarization", model="philschmid/bart-large-cnn-samsum")
+    summary = summarizer(transcript, max_length=256, min_length=60, do_sample=False)
 
-    st.subheader("📋 Summary")
+    st.subheader("📌 Summary")
     st.write(summary[0]["summary_text"])
 
     # Emotion tagging
-    st.info("Analyzing emotional tone...")
+    st.info("🎭 Extracting emotional tones...")
     emotion_model = pipeline("text-classification", model="j-hartmann/emotion-english-distilroberta-base", return_all_scores=True)
-    emotion_results = emotion_model(transcript)
+    emotion_scores = emotion_model(transcript)
 
-    # Aggregate emotions
-    avg_scores = {}
-    for result in emotion_results[0]:
-        avg_scores[result['label']] = round(result['score'] * 100, 2)
-
-    st.subheader("💬 Emotional Insights")
-    for emotion, score in avg_scores.items():
-        st.write(f"{emotion}: {score}%")
+    st.subheader("💬 Emotional Insights (Overall)")
+    for emo in emotion_scores[0]:
+        st.write(f"{emo['label']}: {round(emo['score']*100, 2)}%")
 
     os.remove(audio_path)