🎶 Cover Identification System Using Lyrics Only

📋 Project Overview

This project aims to build a system that takes a YouTube music video URL and returns the title and artist of the song by matching the lyrics with a database of songs. The goal is to identify potential cover songs by analyzing the lyrics.

🛠️ Setup

1. Download a lyrics dataset.
2. Extract embeddings for each song's lyrics using BERT.
3. Create a vector index (database) for fast retrieval of similar lyrics using FAISS.

🔍 Query Process For each YouTube URL (query):

1. Download the YouTube video in a temporary file.
2. Transcribe the lyrics using the Insanely-fast-whisper model.
3. Extract the embeddings of the transcribed lyrics using BERT.
4. Search the top-k similar entries in your vector database using FAISS and return the song title and artist.

🔧 Tools and Technologies

• Insanely-fast-whisper for transcribing lyrics.
• BERT for extracting embeddings.
• FAISS for creating a lyrics index and retrieving similar