Transcribify

Transcribify is a Python package that helps you download audio from YouTube, transcribe it using OpenAI's Whisper, generate embeddings with Sentence Transformers, and store/query data in LanceDB. Perfect for building search-enabled video and audio repositories.

Features

• 🎥 Download Audio: Extract and download audio from YouTube videos.
• 📝 Transcribe: Convert audio files to text with OpenAI's Whisper.
• 🤖 Generate Embeddings: Use Sentence Transformers to create semantic embeddings of transcripts.
• 📦 Store & Query: Store transcripts and embeddings in LanceDB for efficient search and retrieval.
• 🔍 Search Transcripts: Query stored transcripts using semantic similarity.

Installation

Install Transcribify directly from PyPI:

pip install transcribify

Requirements

• Python 3.7 or higher
• Dependencies:
  • whisper
  • sentence-transformers
  • lancedb
  • yt-dlp
  • click

CLI Usage

Transcribify comes with a command-line interface for easy interaction.

1. Download Audio from YouTube

transcribify download <youtube_url> --output_dir <output_directory>

Example:

transcribify download https://www.youtube.com/watch?v=abcd1234 --output_dir downloads

2. Transcribe Audio

transcribify transcribe <audio_file>

Example:

transcribify transcribe downloads/example.mp3

3. Process Audio and Store in LanceDB

transcribify process <audio_file> --db_path <path_to_db> --table_name <table_name>

Example:

transcribify process downloads/example.mp3 --db_path lancedb --table_name transcripts

4. Search Transcripts in LanceDB

transcribify search <query> --db_path <path_to_db> --table_name <table_name>

Example:

transcribify search "What is this video about?" --db_path lancedb --table_name transcripts

Programmatic Usage

You can also use Transcribify programmatically in your Python scripts.

Example Workflow

from transcribify.youtube_downloader import download_audio_from_youtube
from transcribify.transcriber import transcribe_audio
from transcribify.embedder import generate_embeddings
from transcribify.lancedb_manager import LanceDBManager
from transcribify.query_engine import query_lancedb

# Step 1: Download audio from YouTube
audio_path = download_audio_from_youtube("https://www.youtube.com/watch?v=abcd1234")

# Step 2: Transcribe audio
transcript = transcribe_audio(audio_path)

# Step 3: Generate embeddings
embeddings = generate_embeddings([transcript])

# Step 4: Store transcript and embeddings in LanceDB
db_manager = LanceDBManager(db_path="lancedb")
db_manager.insert_data(
    table_name="transcripts",
    data=[{"text": transcript, "embedding": embeddings[0]}]
)

# Step 5: Search for transcripts
results = query_lancedb("What is this video about?", db_path="lancedb", table_name="transcripts")
print(results)

License

This project is licensed under the MIT License. See the LICENSE file for details.

Acknowledgments

• OpenAI for the amazing Whisper transcription model.
• Hugging Face for the Sentence Transformers.
• LanceDB for the efficient database engine for machine learning use cases.
• yt-dlp for YouTube audio/video downloads.