Create speech_input_interim.py

speech_input_interim.py

@@ -0,0 +1,146 @@
+#ref: https://www.youtube.com/watch?v=3ZDVmzlM6Nc
+
+import os
+import chromadb
+import streamlit as st
+from langchain_huggingface import HuggingFaceEmbeddings
+from langchain_chroma import Chroma
+from langchain_groq import ChatGroq
+from langchain.memory import ConversationBufferMemory
+from langchain.chains import ConversationalRetrievalChain
+from PyPDF2 import PdfReader
+from groq import Groq
+from streamlit_webrtc import webrtc_streamer, AudioProcessorBase, WebRtcMode
+import av
+
+# Clear ChromaDB cache to fix tenant issue
+chromadb.api.client.SharedSystemClient.clear_system_cache()
+
+# Ensure required environment variables are set
+GROQ_API_KEY = os.getenv("GROQ_API_KEY")
+if not GROQ_API_KEY:
+    st.error("GROQ_API_KEY is not set. Please configure it in environment variables.")
+    st.stop()
+
+# Initialize Groq Client for transcription and LLM
+groq_client = Groq(api_key=GROQ_API_KEY)
+llm = ChatGroq(model="llama-3.1-70b-versatile", temperature=0, groq_api_key=GROQ_API_KEY)
+
+# Function to process PDFs and set up the vectorstore
+def process_and_store_pdfs(uploaded_files):
+    texts = []
+    for uploaded_file in uploaded_files:
+        reader = PdfReader(uploaded_file)
+        for page in reader.pages:
+            texts.append(page.extract_text())
+
+    embeddings = HuggingFaceEmbeddings()
+    vectorstore = Chroma.from_texts(texts, embedding=embeddings, persist_directory="vector_db_dir")
+    return vectorstore
+
+# Function to set up the chat chain
+def chat_chain(vectorstore):
+    retriever = vectorstore.as_retriever()
+    memory = ConversationBufferMemory(output_key="answer", memory_key="chat_history", return_messages=True)
+
+    chain = ConversationalRetrievalChain.from_llm(
+        llm=llm,
+        retriever=retriever,
+        chain_type="stuff",
+        memory=memory,
+        verbose=True,
+        return_source_documents=True
+    )
+    return chain
+
+# Transcribe audio using Groq Whisper
+def transcribe_audio(file_path):
+    """Transcribe audio using Groq's Whisper model."""
+    with open(file_path, "rb") as file:
+        transcription = groq_client.audio.transcriptions.create(
+            file=(file_path, file.read()),
+            model="distil-whisper-large-v3-en",
+            response_format="json",
+            language="en"
+        )
+    return transcription.text
+
+# Audio Processor Class for Recording
+class AudioProcessor(AudioProcessorBase):
+    def recv(self, frame: av.AudioFrame) -> av.AudioFrame:
+        return frame
+
+# Streamlit UI
+st.title("Chat with PDFs via Speech/Text 🎙️📝📚")
+
+uploaded_files = st.file_uploader("Upload PDF Files", accept_multiple_files=True, type=["pdf"])
+
+if uploaded_files:
+    vectorstore = process_and_store_pdfs(uploaded_files)
+    chain = chat_chain(vectorstore)
+    st.success("PDFs processed! Ready to chat.")
+
+    input_method = st.radio("Choose Input Method", ["Text Input", "Record Audio", "Upload Audio File"])
+
+    # Text Input Mode
+    if input_method == "Text Input":
+        query = st.text_input("Ask your question:")
+        if query:
+            with st.spinner("Thinking..."):
+                response = chain({"question": query})["answer"]
+                st.write(f"**Response:** {response}")
+
+    # Record Audio
+    elif input_method == "Record Audio":
+        st.write("Record your audio query:")
+        webrtc_ctx = webrtc_streamer(
+            key="record",
+            mode=WebRtcMode.SENDONLY,
+            audio_receiver_size=1024,
+            audio_processor_factory=AudioProcessor,
+            media_stream_constraints={"audio": True, "video": False},
+        )
+
+        if webrtc_ctx.audio_receiver:
+            st.write("Recording...")
+            audio_frames = []
+            while True:
+                frame = webrtc_ctx.audio_receiver.recv()
+                audio_frames.append(frame)
+                if len(audio_frames) > 5:  # Stop recording after a few frames
+                    break
+
+            # Save the recorded audio
+            audio_file_path = "recorded_audio.wav"
+            with av.open(audio_file_path, "w") as f:
+                for frame in audio_frames:
+                    f.write(frame)
+            st.success("Recording complete!")
+
+            # Transcribe and Generate Response
+            st.write("Transcribing audio...")
+            transcription = transcribe_audio(audio_file_path)
+            st.write(f"**You said:** {transcription}")
+
+            with st.spinner("Generating response..."):
+                response = chain({"question": transcription})["answer"]
+                st.write(f"**Response:** {response}")
+
+    # Upload Audio File Mode
+    elif input_method == "Upload Audio File":
+        uploaded_audio = st.file_uploader("Upload an audio file (.wav, .mp3)", type=["wav", "mp3"])
+        if uploaded_audio:
+            audio_file_path = "uploaded_audio.wav"
+            with open(audio_file_path, "wb") as f:
+                f.write(uploaded_audio.read())
+
+            st.audio(audio_file_path, format="audio/wav")
+            st.write("Transcribing audio...")
+            transcription = transcribe_audio(audio_file_path)
+            st.write(f"**You said:** {transcription}")
+
+            with st.spinner("Generating response..."):
+                response = chain({"question": transcription})["answer"]
+                st.write(f"**Response:** {response}")
+else:
+    st.info("Please upload PDF files to start chatting.")