ytxt 0.2.0

Local audio and YouTube transcription CLI

ytxt

Local-first, privacy-focused transcription CLI.

ytxt is a developer-centric tool for transcribing audio from YouTube, web URLs, or local files. It bridges the gap between yt-dlp and faster-whisper, providing a seamless, automated pipeline that runs entirely on your machine.

Why ytxt?

• Zero-Cloud Privacy: No data leaves your machine. Perfect for sensitive meetings or private research.
• High Performance: Powered by faster-whisper (CTranslate2), which is up to 4x faster than OpenAI's original implementation.
• Battery Included: Handles downloading, audio extraction (ffmpeg), and transcription in one command.
• Smart Caching: Avoid redundant computations. ytxt hashes inputs to skip re-transcribing files you've already processed.
• Universal: Supports 1,000+ sites via yt-dlp.

Installation

Requires ffmpeg installed on your system.

# Using pip
pip install ytxt

# Using uv (recommended for speed)
uv tool install ytxt

Quick Start (CLI)

Transcribe any YouTube video to Markdown with timestamps:

ytxt "https://www.youtube.com/watch?v=dQw4w9WgXcQ" --format markdown --timestamps --output transcript.md

Power User Tricks

Pipe to an LLM for summarization:

ytxt <url> | llm "Summarize this transcript for a technical audience"

Extract metadata with jq:

ytxt <url> --format json | jq '.[].text'

Library Usage

ytxt is designed to be imported into your own Python automation scripts.

from ytxt import download_audio, transcribe_audio

# 1. Download & Extract
audio_path = download_audio("https://youtube.com/...")

# 2. Transcribe Locally
transcript = transcribe_audio(audio_path, model_size="medium")

# 3. Use the result (list of dicts with 'start', 'end', 'text')
for segment in transcript:
    print(f"[{segment['start']}] {segment['text']}")

Configuration

Option	Description	Default
--model	Whisper model size (tiny, base, small, medium, large-v3)	base
--format	Output format (text, markdown, srt, json)	text
--timestamps	Include timestamps in text/markdown output	False
--no-cache	Force re-transcription by ignoring cache	False

Development

git clone https://github.com/rayanrane/ytxt.git
cd ytxt
pip install -e .

License

MIT