talk2data / app_sound.py
prithvi55's picture
Rename app.py to app_sound.py
252a7bf verified
Raw History Blame Contribute Delete
3.38 kB
import streamlit as st
import pandas as pd
import sqlalchemy
from langchain.schema import Document
from langchain.vectorstores import Chroma
from sentence_transformers import SentenceTransformer
from PyPDF2 import PdfReader
import torch
import tempfile
import os
import numpy as np
import matplotlib.pyplot as plt
import plotly.express as px
import openpyxl
import speech_recognition as sr
from gtts import gTTS
import base64
from streamlit_mic_recorder import mic_recorder
# ------------------
# CONFIG
# ------------------
CHROMA_DIR = "chromadb_local"
COLLECTION_NAME = "prompts"
EMBEDDING_MODEL = "all-MiniLM-L6-v2"
# ------------------
# CHROMA DB FUNCTIONS
# ------------------
def get_chroma_client():
from chromadb import Client
from chromadb.config import Settings
return Client(Settings(persist_directory=CHROMA_DIR))
def get_chroma_collection(client):
try:
return client.get_collection(COLLECTION_NAME)
except Exception:
return client.create_collection(COLLECTION_NAME)
def get_embedding_model():
device = "cuda" if torch.cuda.is_available() else "cpu"
cache_dir = tempfile.mkdtemp()
return SentenceTransformer(EMBEDDING_MODEL, device=device, cache_folder=cache_dir)
# ------------------
# SPEECH TO TEXT
# ------------------
def speech_to_text(audio_bytes):
recognizer = sr.Recognizer()
with tempfile.NamedTemporaryFile(delete=False, suffix=".wav") as tmp_wav:
tmp_wav.write(audio_bytes)
tmp_wav.flush()
with sr.AudioFile(tmp_wav.name) as source:
audio_data = recognizer.record(source)
try:
text = recognizer.recognize_google(audio_data)
return text
except sr.UnknownValueError:
return "Sorry, could not understand the audio."
except sr.RequestError:
return "Speech Recognition service error."
# ------------------
# MAIN APP
# ------------------
st.set_page_config(page_title="Talk2Data - MultiModal RAG", layout="wide")
st.title("๐ŸŽ™๏ธ Talk2Data โ€“ MultiModal RAG")
# Mic Recorder
col1, col2 = st.columns([2, 1])
with col1:
user_query = st.text_input("๐Ÿ’ฌ Enter your query:")
with col2:
st.write("๐ŸŽค Or speak your query")
audio = mic_recorder(
start_prompt="๐ŸŽ™๏ธ Start Recording",
stop_prompt="โน Stop Recording",
just_once=True,
key="voice"
)
if audio:
st.audio(audio["bytes"])
transcribed_text = speech_to_text(audio["bytes"])
st.success(f"**Transcribed Query:** {transcribed_text}")
if not user_query:
user_query = transcribed_text # If no typed query, use voice
# RAG Process
if st.button("Run Query"):
if user_query:
st.info(f"Processing: {user_query}")
# Here you integrate with your existing SQL + Vector Search pipeline
# Example:
client = get_chroma_client()
collection = get_chroma_collection(client)
model = get_embedding_model()
# Vector embedding
query_embedding = model.encode([user_query])[0]
# Dummy search (replace with actual retrieval)
results = collection.query(
query_embeddings=[query_embedding],
n_results=5
)
st.write("### Results")
st.write(results)
else:
st.warning("Please type or speak your query first.")