ytpipe 0.1.7

YouTube audio download and faster-whisper transcription pipeline.

ytpipe

YouTube audio download and transcription pipeline.

ytpipe downloads audio from YouTube (videos, playlists, channels) and transcribes locally using faster-whisper. Designed for practical, repeatable ingestion of YouTube content into clean on-disk datasets.

Features

• One-command pipeline: download + transcribe in one step
• Robust downloads powered by yt-dlp (videos, playlists, channels)
• Fast local transcription via faster-whisper (CPU or NVIDIA GPU)
• VAD enabled by default for better quality and faster processing
• Multiple output formats: JSON, TXT, SRT, VTT subtitles
• Parallel processing for batch transcription (multi-CPU and multi-GPU)

Requirements

• Python >= 3.10
• FFmpeg (must be in PATH or provided via --ffmpeg)
• Optional: NVIDIA GPU + CUDA for acceleration

Installation

CPU only (default)

pip install ytpipe

GPU support (NVIDIA CUDA)

pip install "ytpipe[gpu]"

Quick Start

Full pipeline (download + transcribe)

# CPU (default)
ytpipe pipeline -s "https://www.youtube.com/watch?v=VIDEO_ID" --out data

# GPU accelerated
ytpipe pipeline -s "https://www.youtube.com/watch?v=VIDEO_ID" --out data --device cuda

Download only

ytpipe download -s "https://www.youtube.com/watch?v=VIDEO_ID" --out data/raw

Download from channel (with limit)

# Download 5 videos from a channel
ytpipe pipeline -s "https://www.youtube.com/@ChannelName" --out data --max 5

# Or download only (without transcription)
ytpipe download -s "https://www.youtube.com/@ChannelName" --out data/raw --max 10

Transcribe only

ytpipe transcribe --audio-dir data/raw/audio --out data/transcripts

Output Structure

data/
  raw/
    audio/
      VIDEO_ID.m4a
      VIDEO_ID.info.json
    manifest.jsonl
    downloaded.txt
  transcripts/
    VIDEO_ID.json
    VIDEO_ID.txt

CLI Options

Common options

Option	Description
--model	Whisper model: small, medium, large-v3 (default: medium)
--device	Device: cpu, cuda, auto (default: cpu)
--language	Language hint (e.g., en). Auto-detects if not set
--output-format	Output formats: json,txt,srt,vtt (default: json,txt)

GPU acceleration

ytpipe transcribe --audio-dir data/raw/audio --device cuda --compute-type float16

Parallel processing

# CPU parallel (4 workers)
ytpipe transcribe --audio-dir data/raw/audio --workers 4

# Multi-GPU
ytpipe transcribe --audio-dir data/raw/audio --device cuda --workers 2

Python API

from pathlib import Path
from ytpipe import (
    DownloadConfig,
    TranscriptionConfig,
    download_sources,
    transcribe_directory,
)

# Download
items = download_sources(
    sources=["https://www.youtube.com/watch?v=VIDEO_ID"],
    config=DownloadConfig(out_dir=Path("data/raw")),
)

# Transcribe
results = transcribe_directory(
    audio_dir=Path("data/raw/audio"),
    out_dir=Path("data/transcripts"),
    config=TranscriptionConfig(model="medium"),
)

Troubleshooting

• FFmpeg not found: Install FFmpeg and add to PATH, or use --ffmpeg /path/to/ffmpeg
• GPU not detected: Ensure NVIDIA drivers are installed and nvidia-smi works
• DLL errors on Windows: Use --device cpu (default) to avoid CUDA dependencies

License

MIT License