Upload 4 files

.env

@@ -0,0 +1,4 @@
+GROQ_API_KEY="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"
+RAPIDAPI_LANG_TRANS="0875969c18mshea4b5d106303222p197cb2jsnd47e91b7da30"
+RAPIDAPI_HOST="microsoft-translator-text.p.rapidapi.com"
+TAVILY_API_KEY="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"

app.py

@@ -0,0 +1,81 @@
+import streamlit as st
+from dotenv import load_dotenv
+from audiorecorder import audiorecorder
+from langchain_core.messages import HumanMessage, AIMessage
+import requests
+from transformers import pipeline
+from gtts import gTTS
+import io
+
+# Load environment variables (if any)
+load_dotenv()
+
+user_id = "1"  # example user id
+
+# Initialize the wav2vec2 model for Urdu speech-to-text
+pipe = pipeline("automatic-speech-recognition", model="kingabzpro/wav2vec2-large-xls-r-300m-Urdu")
+
+def get_response(user_input):
+    '''
+        Takes user_input in English and invokes the infer API for response.
+
+        Parameters:
+            user_input (string): User Query in English.
+        Returns:
+            res (string): Response from the LLM.
+    '''
+    url = f"http://127.0.0.1/infer/{user_id}"
+    headers = {"Content-Type": "application/x-www-form-urlencoded"}
+    data = {"user_input": user_input}
+    response = requests.post(url, headers=headers, data=data)
+    res = response.json()
+    return res["data"]
+
+
+def text_to_speech(text, lang='ur'):
+    '''
+        Converts text to speech using gTTS.
+
+        Parameters:
+            text (string): Text to be converted to speech.
+            lang (string): Language for the speech synthesis. Default is 'ur' (Urdu).
+        Returns:
+            response_audio_io (BytesIO): BytesIO object containing the audio data.
+    '''
+    tts = gTTS(text, lang=lang)
+    response_audio_io = io.BytesIO()
+    tts.write_to_fp(response_audio_io)
+    response_audio_io.seek(0)
+    return response_audio_io
+
+
+st.set_page_config(page_title="Urdu Virtual Assistant", page_icon="🤖")  # set the page title and icon
+st.title("Urdu Virtual Assistant")  # set the main title of the application
+
+# Add a text input box
+audio = audiorecorder()
+
+if len(audio) > 0:
+    # Save the audio to a file
+    audio.export("audio.wav", format="wav")
+
+    # Convert audio to text using the wav2vec2 model
+    with open("audio.wav", "rb") as f:
+        audio_bytes = f.read()
+
+    # Process the audio file
+    result = pipe("audio.wav")
+    user_query = result["text"]
+
+    with st.chat_message("Human"):  # create the message box for human input
+        st.audio(audio.export().read())  # display the audio player
+        st.markdown(user_query)
+
+    # Get response from the LLM
+    response_text = get_response(user_input=user_query)
+    response_audio = text_to_speech(response_text, lang='ur')
+
+    # Play the generated speech in the app
+    with st.chat_message("AI"):
+      st.audio(response_audio.read(), format='audio/mp3')
+      st.markdown(response_text)

llm.py

@@ -0,0 +1,99 @@
+from fastapi import FastAPI, Form
+from langchain_core.runnables.base import RunnableSequence
+from langchain_core.prompts import ChatPromptTemplate
+from langchain_groq import ChatGroq
+import os
+import requests
+from dotenv import load_dotenv
+from langgraph.checkpoint.memory import MemorySaver
+from langgraph.prebuilt import create_react_agent
+from langchain_community.tools.tavily_search import TavilySearchResults
+load_dotenv()
+
+app = FastAPI()
+
+llm = ChatGroq(
+    model="llama-3.1-70b-versatile",
+    temperature=0,
+    max_tokens=None,
+    timeout=None,
+    max_retries=5,
+    groq_api_key=os.getenv("GROQ_API_KEY")
+)
+
+search = TavilySearchResults(
+      max_results=2,
+    )
+tools = [search]
+memory = MemorySaver()
+
+agent_executor = create_react_agent(llm, tools, checkpointer=memory)
+
+def translate(target, text):
+  '''
+    Translates given text into target language
+
+    Parameters:
+        target (string): 2 character code to specify the target language.
+        text (string): Text to be translated.
+
+    Returns:
+        res (string): Translated text.
+  '''
+  url = "https://microsoft-translator-text.p.rapidapi.com/translate"
+
+  querystring = {"api-version":"3.0","profanityAction":"NoAction","textType":"plain", "to":target}
+
+  payload = [{ "Text": text }]
+  headers = {
+    "x-rapidapi-key": os.getenv("RAPIDAPI_LANG_TRANS"),
+    "x-rapidapi-host": "microsoft-translator-text.p.rapidapi.com",
+    "Content-Type": "application/json"
+  }
+
+  response = requests.post(url, json=payload, headers=headers, params=querystring)
+  res = response.json()
+  return res[0]["translations"][0]["text"]
+
+
+@app.post('/infer/{user_id}')
+def infer(user_id: str, user_input: str = Form(...)):
+    '''
+        Returns the translated response from the LLM in response to a user query.
+
+        Parameters:
+            user_id (string): User ID of a user.
+            user_input (string): User query.
+
+        Returns:
+            JSON Response (Dictionary): Returns a translated response from the LLM.
+    '''
+
+    user_input = translate("en", user_input) # translate user query to english
+
+    prompt = ChatPromptTemplate.from_messages( # define a prompt
+        [
+            (
+                "system",
+                "You're a compassionate AI virtual Assistant"
+            ),
+            ("human", "{user_input}")
+        ]
+    )
+
+    runnable = prompt | agent_executor # define a chain
+
+    conversation = RunnableSequence( # wrap the chain along with chat history and user input
+        runnable,
+    )
+
+    response = conversation.invoke( # invoke the chain by giving the user input and the chat history
+        {"user_input": user_input},
+        config={"configurable": {"thread_id":user_id}}
+    )
+
+    res = translate("ur", response["messages"][-1].content) # translate the response to Urdu
+
+    return {
+        "data": res
+    }

requirements.txt

@@ -0,0 +1,20 @@
+langgraph
+streamlit-audiorecorder
+fastapi
+langchain
+python-dotenv
+uvicorn
+gTTS
+openai-whisper
+langchain_core
+langchain_community
+langchain_groq
+python-multipart
+kenlm
+pyctcdecode
+transformers
+librosa
+soundfile
+gTTS
+tavily-python
+langgraph-checkpoint-sqlite