tutorial97

tutorial97/main.py

@@ -0,0 +1,71 @@
+import streamlit as st
+from pytube import YouTube
+import os
+from split_mp3 import split_audio
+from transcript_generator import generate_transcript
+from langchain_community.document_loaders import TextLoader
+from langchain_openai import ChatOpenAI,OpenAIEmbeddings
+from langchain_chroma import Chroma
+from langchain.text_splitter import RecursiveCharacterTextSplitter
+from langchain.chains import RetrievalQA
+
+# Initialize session state variables
+if 'step' not in st.session_state:
+    st.session_state.step = 0
+if 'input_url' not in st.session_state:
+    st.session_state.input_url = ""
+if 'qa_chain' not in st.session_state:
+    st.session_state.qa_chain = None
+
+cwd = os.getcwd()
+
+st.title("YouTube Video Chat App")
+
+if st.session_state.step == 0:
+    st.write("Enter your video URL:")
+    st.session_state.input_url = st.text_area("Enter your video URL")
+    if st.button("Submit"):
+        if st.session_state.input_url:
+            st.session_state.step = 1
+            st.experimental_rerun()
+
+if st.session_state.step == 1:
+    st.video(st.session_state.input_url)
+    yt = YouTube(st.session_state.input_url)
+    data = yt.streams.filter(only_audio=True).first()
+    output_file = data.download(output_path=cwd)
+    os.rename(output_file, "sample.mp3")
+
+    input_audio_path = os.path.join(cwd, "sample.mp3")
+    output_folder = cwd
+    segment_length = 600000  # Segment length for splitting audio
+
+    split_audio(input_audio_path, output_folder, segment_length)
+    st.success("All MP3s are split successfully")
+
+    generate_transcript()
+    st.success("Transcript file generated successfully")
+
+    loader = TextLoader(os.path.join(cwd, "transcript.txt"))
+    docs = loader.load()
+
+    text_splitter = RecursiveCharacterTextSplitter(chunk_size=5000, chunk_overlap=150)
+    splits = text_splitter.split_text(docs[0].page_content)
+
+    embeddings = OpenAIEmbeddings()
+    vectordb = Chroma.from_texts(splits, embedding=embeddings, persist_directory=os.path.join(cwd, "output"))
+    vs=Chroma(persist_directory=os.path.join(cwd, "output"),embedding_function=embeddings)
+    st.session_state.qa_chain = RetrievalQA.from_chain_type(
+        llm=ChatOpenAI(model="gpt-4"),
+        chain_type="stuff",
+        retriever=vectordb.as_retriever()
+    )
+
+    st.session_state.step = 2
+    st.experimental_rerun()
+
+if st.session_state.step == 2:
+    input_qsn = st.text_input("Ask your question")
+    if st.button("Submit Question"):
+        response = st.session_state.qa_chain.invoke(input_qsn)
+        st.write(response["result"])