SentioVoice

SentioVoice is an application that allows users to submit feedback, analyzes sentiment using Azure AI Services, and automatically generates responses (in both text and audio form) using a Large Language Model.

Table of Contents

• SentioVoice
  • Table of Contents
  • Features
  • Demo
    • Response to a positive feedback
    • Response to a neutral feedback
    • Response to a negative feedback
  • Component breakdown
    • Frontend (Streamlit)
    • API (Azure Functions + FastAPI)
    • Azure Cognitive Services
    • LLM Azure OpenAI
  • Setup instructions
    • Development environment
      • Prerequisites
      • Setup steps
        • Install the dependencies
        • Configure environment variables
          • Environment variables description
        • Run the application
    • Running with Docker
      • Configure environment
      • Bring the containers up
  • Usage

---

Features

• Feedback Submission: Users can submit free-text feedback via a user-friendly interface.
• Sentiment Analysis: Utilizes Azure Text Analytics API to determine the sentiment (positive, negative, or neutral) of the submitted feedback.
• Automated Response Generation: Uses an LLM (GPT-3 or GPT-4) to generate a concise, context-aware response based on the analyzed sentiment.
• Audio Response: Converts the generated text response into audio using Azure Text-to-Speech (TTS) services.
• Interactive Dashboard: Provides a simple UI to display:
  • The submitted feedback.
  • The sentiment analysis results.
  • The generated text response.
  • An option to play or download the audio file of the response.

---

Demo

Response to a positive feedback

Response to a neutral feedback

Response to a negative feedback

---

Component breakdown

Frontend (Streamlit)

• Receives feedback from users.
• Displays the sentiment, text response, and audio controls.

API (Azure Functions + FastAPI)

• Processes the feedback.
• Interfaces with Azure Cognitive Services for sentiment analysis.
• Calls the LLM API to generate responses.
• Converts text responses to audio using Azure TTS.

Azure Cognitive Services

Provides both sentiment analysis and text-to-speech capabilities.

LLM Azure OpenAI

Generates contextual responses based on the feedback and sentiment.

---

Setup instructions

Development environment

Prerequisites

• Python 3.11
• Poetry ^2.0
• Azure Functions Core Tools
• An Azure account with the following services provisioned:
  • Azure AI Services: - Text Analytics - Speech synthesis
  • Azure OpenAI service provisioned with a LLM model

Setup steps

Install the dependencies

Navigate to each project subdirectory and install the dependencies:

cd api
poetry install --with dev

cd ../ui
poetry install --with dev

Configure environment variables

Configure them based on the .env.example files:

cd api
cp .env.example .env
vim .env # Fill the variables

cd ../ui
cp .env.example .env
vim .env # Fill the variables

Environment variables description

Streamlit:

Variable	Description
API_URL	SentioVoice API URL.
AUDIO_PATH	Downloaded Audio temp path.

API

Variable	Description
AZURE_OPENAI_API_KEY	API key for Azure OpenAI service.
AZURE_OPENAI_DEPLOYMENT	Deployed OpenAI model name.
AZURE_OPENAI_MODEL	OpenAI model (e.g., gpt-4o-mini).
AZURE_OPENAI_API_VERSION	Azure OpenAI API version
AZURE_OPENAI_ENDPOINT	Azure OpenAI endpoint URL.
AZURE_OPENAI_TEMPERATURE	Model temperature (e.g.; 0.7).
AZURE_OPENAI_MAX_TOKENS	Maximum tokens for responses. (e.g.; 50)
AZURE_OPENAI_SYSTEM_PROMPT	Initial system prompt for the model.
AZURE_AI_SERVICES_API_KEY	API key for Azure AI Services.
AZURE_AI_SERVICES_ENDPOINT	Azure AI Services endpoint URL.
AZURE_AI_SERVICES_REGION	Region for Azure AI Services. (e.g.; westeurope)
AZURE_AI_SERVICES_AUDIO_PATH	Path for storing audio files. (e.g.; audio)
PROMPT_FILE	Path to the prompt configuration file. (e.g.; prompts)

Run the application

Start the Azure functions host:

cd api
func start

This will launch the API at http://localhost:7071

In another terminal window, start the Streamlit app:

cd ui
streamlit run app.py

This will launch the UI at http://localhost:8501

Running with Docker

The current implementation does not support setting up a development environment (i.e.; no volumes, etc.). It is intended to be used as a deployment emulation to showcase the application.

Configure environment

Configure the environment in the same way as described above, with a slight difference:

# ui/.env
API_URL=http://api/

This is because a network has been defined in the docker-compose.yml file and the API is served there as api(i.e.; service_name):80 (default Azure Functions image configuration).

Bring the containers up

docker-compose up --build -d

The services will be launched in the same ports as the development environment:

• API: http://localhost:7071 (useful to check the API documentation)
• UI: http://localhost:8501 This will launch the API at http://localhost:7071 This will launch the UI at http://localhost:8501

---

Usage

1. Open the Frontend: Navigate to the Streamlit UI (http://localhost:8501).
2. Submit Feedback: Enter your feedback in the provided text box, leave the form focus and click Submit feedback button.
3. Review the Results:
   • The sentiment of your feedback is analyzed and displayed.
   • A text response generated by the LLM is shown.
   • An audio version of the response is available.
     • Audio options:
       • Play: Listen to the response directly in the UI.
       • Download: Click the download button in the player menu to download the generated audio.