🎵 Shazam Clone : Audio Transcription & Subtitle Search

Overview

This project focuses on building AI-powered search systems, audio-to-text conversion, and interactive search applications. It integrates search engine technologies, vector embeddings, speech-to-text models, and subtitle processing to enable efficient information retrieval and media analysis.

Deployment URL :

Shazam Clone Search Engine

• Note - URL is available for 72 hours only.

Objective

Develop an advanced search engine algorithm that efficiently retrieves subtitles based on user queries, leveraging natural language processing (NLP) and machine learning to enhance search relevance and accuracy.

• Keyword-based vs. Semantic Search Engines
• Keyword-Based Search Engine: Relies on exact keyword matches between user queries and indexed documents.
• Semantic Search Engine: Understands the meaning and context of queries and documents beyond keyword matching.

Core Logic

The search engine compares user queries against subtitle documents through these key steps:

1. Preprocessing the Data

• Extract and clean subtitle data (remove timestamps, punctuation, etc.).

• Subset the dataset if compute resources are limited (e.g., use 30% of the data).

2. Generating Text Vectors

• Convert subtitle text into vector embeddings using different techniques:

• BOW (Bag of Words) / TF-IDF → For keyword-based search.

• BERT-based SentenceTransformers → For semantic search.

3. Implementing a Document Chunker

• Break down large subtitles into smaller chunks to preserve context.

• Use overlapping windows to prevent information loss.

4. Storing Embeddings in ChromaDB

• Store vector representations of subtitle chunks in ChromaDB for efficient retrieval.

5. Retrieving Documents Based on User Query

• Convert user audio query to text (using AssemblyAI or other ASR models).
• Vectorize the text query and compute cosine similarity between query and subtitle embeddings.
• Rank and return the most relevant subtitle segments.

Notebooks & Descriptions

1. Audio_2_Text.ipynb

• Converts audio files into text using speech recognition models using AssemblyAI.
• Useful for transcription services, podcasts, and accessibility applications.

2. Chroma_db_Embeddings_V2.ipynb

• Uses ChromaDB to create and manage vector embeddings.
• Enables efficient similarity search, semantic retrieval, and document indexing.
• Applied in question-answering systems, AI-powered search engines, and NLP tasks.

3. Gradio_Search_Engine_Demo.ipynb

• Implements a search engine demo using Gradio, allowing users to test queries interactively.
• Supports keyword-based search, vector search (semantic search), or hybrid models.
• Likely integrates with ChromaDB embeddings for intelligent retrieval.

4. Search_Engine_Extracting_Data.ipynb

• Focuses on data extraction, scraping, and indexing for a search engine.
• Handles text preprocessing, tokenization, and document structuring.
• Useful for web crawlers, automated indexing, and structured data retrieval.

5. Shazam_Clone_Search_Engine.ipynb

• Builds a music recognition system similar to Shazam.
• Enables identification of audio, speeches, or sound patterns from audio samples.

6. Subtitles_Chunking.ipynb

• Processes subtitle files by splitting them into smaller text segments.
• Helps in automatic subtitle generation, video search indexing, and multilingual translation.
• Includes time-stamping, sentence segmentation, and embedding generation for efficient retrieval.

7. Testing_Search_Mechanism.ipynb

• Evaluates the effectiveness of different search algorithms.
• Tests ranking models, BM25, TF-IDF, and vector-based retrieval methods.

Libraries Used :

• gradio
• assemblyai
• pydub
• python-dotenv
• chromadb
• sentence-transformers

Applications

• Multimodal Search Engines: Combining text, audio, and embeddings for intelligent retrieval.
• Audio & Music Recognition: Implementing a Shazam-like system for identifying songs and sound patterns.
• Semantic Search & NLP: Using vector embeddings for AI-driven document retrieval.
• Interactive Demos & UI: Leveraging Gradio for easy-to-use search engine interfaces.
• Video & Subtitle Processing: Extracting and structuring subtitles for indexing and accessibility.

Summary

This project integrates cutting-edge AI search technologies, text/audio processing, and interactive search tools. With vector databases, NLP models, and search ranking algorithms, it offers advanced media analysis, transcription, and retrieval capabilities. 🚀

---

Want to Contribute?

Feel free to open issues, suggest improvements, or contribute to enhance this project!

Let me know if you need modifications! 😊