app.py

import streamlit as st
import whisper
from transformers import pipeline
import spacy
from summa import keywords
import datetime
import os

@st.cache_resource
def load_models():
    whisper_model = whisper.load_model("base")  # You can use 'small' or 'medium' for better results
    summarizer = pipeline("summarization", model="facebook/bart-large-cnn")
    nlp = spacy.load("en_core_web_sm")
    return whisper_model, summarizer, nlp

def extract_action_items(text, nlp):
    doc = nlp(text)
    actions = []
    
    for sent in doc.sents:
        for token in sent:
            if token.dep_ == "ROOT" and token.pos_ == "VERB":
                action = {
                    "text": sent.text,
                    "responsible": [],
                    "deadline": []
                }
                
                for ent in sent.ents:
                    if ent.label_ == "PERSON":
                        action["responsible"].append(ent.text)
                    elif ent.label_ == "DATE":
                        action["deadline"].append(ent.text)
                
                actions.append(action)
                break
    return actions

def main():
    st.title("🤖 Smart AI Meeting Assistant")
    
    whisper_model, summarizer, nlp = 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)
            
            # Transcribe
            result = whisper_model.transcribe(file_path)
            transcript = result["text"]
        
        st.write(transcript)
        os.remove(file_path)  # Cleanup
        
        st.subheader("Meeting Summary")
        with st.spinner("Generating summary..."):
            truncated_text = transcript[:1024]
            summary = summarizer(truncated_text, max_length=150, min_length=50)[0]['summary_text']
        st.write(summary)
        
        st.subheader("🚀 Action Items")
        actions = extract_action_items(transcript, nlp)
        
        if not actions:
            st.write("No action items detected")
        else:
            for i, action in enumerate(actions, 1):
                responsible = ", ".join(action["responsible"]) or "Unassigned"
                deadline = ", ".join(action["deadline"]) or "No deadline"
                st.markdown(f"""
                **Action {i}**
                - Task: {action["text"]}
                - Responsible: {responsible}
                - Deadline: {deadline}
                """)
        
        st.subheader("🔑 Key Terms")
        key_phrases_result = keywords.keywords(transcript) or ""
        key_phrases = [kp.strip() for kp in key_phrases_result.split("\n") if kp.strip()]
        st.write(", ".join(key_phrases) if key_phrases else "No key terms extracted")

if __name__ == "__main__":
    main()