Enhance EmpathyBot with Relevant Features -- 2

Project Description:

I developed EmpathyBot, a web application built with Streamlit that detects emotions in user text (happy or sad) and generates empathetic responses. The system uses advanced NLP models, embeddings, and FAISS for semantic similarity retrieval to provide meaningful, context-aware replies.

It now also supports audio input: users can record or upload a voice message, which is transcribed, analyzed, and answered empathetically.

Key Features:

Emotion Detection: Uses BERT-based sentiment analysis (nlptown/bert-base-multilingual-uncased-sentiment).
Confidence Visualization: Shows model confidence in predictions.
Semantic Similarity: Retrieves up to 3 similar texts from the dataset using SentenceTransformer embeddings and FAISS.
Empathetic Responses: Synthesizes answers from similar texts to provide emotional support.
Audio Support: Upload or record audio; converts speech to text, predicts emotion, and generates responses with TTS.
Professional UI/UX: Centered layout, highlighted keywords, and bullet-pointed outputs.

Technical Details:

Models:

Sentiment: nlptown/bert-base-multilingual-uncased-sentiment
Embeddings: all-MiniLM-L6-v2
Speech-to-Text: facebook/wav2vec2-base-960h
Dataset: cardiffnlp/tweet_eval (Emotion subset)
Tools: Streamlit, FAISS, SentenceTransformer

Demo & Screenshots:

View Demo on Google Drive : https://drive.google.com/drive/folders/1-sMI0HkxM1E31fvujCetamvZm8MA_Sc1?hl=ar

Use Cases:

Mental wellness chatbots
Customer support with empathetic AI
Social media or survey sentiment analysis

Installation (Optional for Freelancers to Run Locally):

git clone https://github.com/ahmdeltoky03/EmpathyBot.git
cd EmpathyBot
pip install -r requirements.txt
streamlit run app.py


Contact / Author:

Ahmed Eltokhy
GitHub: ahmdeltoky03