Initial commit with Streamlit app and requirements

app.py

@@ -1,49 +1,37 @@
 import streamlit as st
 from transformers import pipeline
-import soundfile as sf
-import torch
+from io import BytesIO
 
-# Load Transcriber model optimized for CPU
+# Load models optimized for CPU
 transcriber = pipeline("automatic-speech-recognition", model="openai/whisper-tiny.en", device=-1)
-
-# Load Summary Model optimized for CPU
 summarizer = pipeline("summarization", model="sshleifer/distilbart-cnn-12-6", device=-1)
+question_generator = pipeline("text2text-generation", model="google/t5-efficient-tiny", device=-1)
 
-# Streamlit file uploader
-uploaded_file = st.file_uploader("Upload an audio file", type=["wav"])
+# Streamlit UI
+st.title("Curate AI - Audio Transcription and Summarization")
 
-# Process the uploaded file
+uploaded_file = st.file_uploader("Upload an audio file", type=["wav", "mp3", "m4a"])
 if uploaded_file is not None:
-    # Save the uploaded file temporarily
-    audio_data, samplerate = sf.read(uploaded_file)
+    st.audio(uploaded_file, format='audio/wav')
+
+    # Convert the uploaded file to a format suitable for the transcription model
+    audio_bytes = BytesIO(uploaded_file.read())
     
     # Transcribing the audio
-    with st.spinner('Transcribing the audio...'):
-        lecture_text = transcriber(uploaded_file)["text"]
-
-    # Preprocessing text
-    num_words = len(lecture_text.split())
-    max_length = min(num_words, 1024)  # BART model max input length is 1024 tokens
-    max_length = int(max_length * 0.75)  # Approx token conversion
-
+    st.write("Transcribing the audio...")
+    lecture_text = transcriber(audio_bytes)["text"]
+    st.write("Transcription: ", lecture_text)
+    
     # Summarization
-    with st.spinner('Summarizing the lecture...'):
-        summary = summarizer(
-            lecture_text,
-            max_length=1024,  # DistilBART max input length
-            min_length=int(max_length * 0.1),
-            truncation=True
-        )
-
-    # Clean up the summary text
-    if not summary[0]["summary_text"].endswith((".", "!", "?")):
-        last_period_index = summary[0]["summary_text"].rfind(".")
-        if last_period_index != -1:
-            summary[0]["summary_text"] = summary[0]["summary_text"][:last_period_index + 1]
-
-    # Output summary
-    st.write("\n### Summary:\n", summary[0]["summary_text"])
-
-else:
-    st.warning("Please upload a valid audio file.")
-
+    st.write("Summarizing the transcription...")
+    num_words = len(lecture_text.split())
+    max_length = min(num_words, 1024)  # Max input for BART is 1024 tokens
+    summary = summarizer(lecture_text, max_length=1024, min_length=int(max_length * 0.1), truncation=True)
+    st.write("Summary: ", summary[0]['summary_text'])
+    
+    # Question Generation
+    context = f"Based on the following lecture summary: {summary[0]['summary_text']}, generate some relevant practice questions."
+    st.write("Generating questions...")
+    questions = question_generator(context, max_new_tokens=50)
+    for question in questions:
+        st.write(question["generated_text"])

requirements.txt

@@ -1,4 +1,2 @@
-transformers
 streamlit
-soundfile
-torch
+transformers