SpeechEcho — Real-Time Voice Cloning & Speech Synthesis
Developed a real-time voice cloning and conversational speech synthesis system using PyTorch and XTTS-v2, with optimized preprocessing and inference pipelines for low-latency, high-quality synthesis. Trained on a custom South Asian accent dataset to help South Asian callers better understand customer support voices during phone conversations, addressing accent barriers and improving communication clarity.
- PyTorch
- XTTS-v2
- Real-Time Inference
Retrieval-Augmented Generation System
LangChain, FAISS
Built an end-to-end RAG pipeline for research-paper-based question answering using LLMs, integrating a FAISS vector database for efficient semantic search and retrieval. Used Grafana and Prometheus to monitor query accuracy and analyze hallucination rates.
Plant Disease Detection
CNN, VGG16
Implemented transfer learning using VGG16 with data augmentation for accurate plant disease classification, and evaluated performance using precision, recall, and confusion matrix analysis.
- TensorFlow
- CNN
- Computer Vision
New York Housing Price Prediction
Performed data cleaning, feature engineering, and exploratory data analysis on the New York housing dataset, building and comparing regression models to predict housing prices with rigorous performance evaluation.
- Data Analysis
- Regression
- Python