Audio Extraction Analysis

Transform video and audio recordings into structured, actionable documentation. Supports multiple transcription providers (Deepgram, ElevenLabs, Whisper, Parakeet) with speaker diarization, topic detection, and sentiment analysis.

Key Features

• Multi-Provider Support — Cloud (Deepgram, ElevenLabs) and local (Whisper, Parakeet) transcription
• URL Ingestion — Direct processing from YouTube, Vimeo, and other platforms
• Interactive TUI — Terminal interface with live progress and health monitoring
• Intelligent Analysis — Speaker separation, topic extraction, sentiment analysis
• Production Ready — Circuit breaker pattern, health checks, path sanitization

Installation

Prerequisites

• Python 3.11+ (3.12 recommended)
• FFmpeg
• API key for cloud providers OR local models (Whisper/Parakeet)

Setup

# Clone and install
git clone https://github.com/edlsh/audio-extraction-analysis.git
cd audio-extraction-analysis
uv sync

# Install FFmpeg
brew install ffmpeg        # macOS
sudo apt install ffmpeg    # Ubuntu
choco install ffmpeg       # Windows

Optional Extras

uv sync --extra tui                    # Terminal UI
uv add openai-whisper torch            # Local Whisper
uv sync --extra parakeet               # NVIDIA Parakeet
uv sync --dev --extra tui --extra parakeet  # All features

Configure Provider

# Cloud: Set API key
export DEEPGRAM_API_KEY='your-key'     # From console.deepgram.com

# Or local: Whisper works without API keys after installation

Verify

audio-extraction-analysis --version

Quick Start

# Process a video file
audio-extraction-analysis process meeting.mp4

# Process from URL
audio-extraction-analysis process --url "https://youtube.com/watch?v=..."

# With custom output
audio-extraction-analysis process video.mp4 --output-dir ./results

# Full analysis (5 files)
audio-extraction-analysis process video.mp4 --analysis-style full

CLI Reference

Commands

Command	Description
process	Full pipeline: extract → transcribe → analyze
extract	Audio extraction only
transcribe	Transcription only
export-markdown	Export transcript as Markdown
tui	Launch interactive terminal UI

Common Options

Option	Values	Description
--quality	speech, standard, high, compressed	Audio quality preset
--language	en, es, fr, de, auto	Transcription language
--provider	auto, deepgram, elevenlabs, whisper, parakeet	Provider selection
--output-dir	Path	Output directory
--analysis-style	concise, full	Single file vs 5-file output
--verbose	Flag	Detailed logging

Examples

# High-quality transcription
audio-extraction-analysis process interview.mp4 --quality high --language en

# Extract audio only
audio-extraction-analysis extract presentation.mp4 --quality speech

# Transcribe existing audio
audio-extraction-analysis transcribe recording.mp3 --provider whisper

# Batch processing
for video in *.mp4; do
  audio-extraction-analysis process "$video" --output-dir "./results/${video%.*}"
done

Interactive TUI

Launch a guided interface with real-time progress monitoring:

audio-extraction-analysis tui
audio-extraction-analysis tui --input video.mp4 --output-dir ./results

Features

• Live progress bars with ETAs
• Color-coded log streaming
• Provider health monitoring
• File browser with recent files
• Dark/light theme toggle
• Auto-saved configuration

Keyboard Shortcuts

Key	Action
q	Quit
d	Toggle dark mode
? / h	Help
c	Cancel pipeline (run screen)
o	Open output folder (when complete)

Output Structure

Concise Mode (default)

./output/
├── meeting.mp3                 # Extracted audio
├── meeting_analysis.md         # Single comprehensive analysis
└── meeting_transcript.txt      # Provider-formatted transcript

Full Mode (--analysis-style full)

./output/
├── meeting.mp3
├── 01_executive_summary.md       # High-level overview
├── 02_chapter_overview.md        # Content breakdown by topic
├── 03_key_topics_and_intents.md  # Technical analysis
├── 04_full_transcript_with_timestamps.md
└── 05_key_insights_and_takeaways.md

Configuration

Environment Variables

# Cloud providers
export DEEPGRAM_API_KEY='...'      # console.deepgram.com
export ELEVENLABS_API_KEY='...'    # elevenlabs.io/api

# Local providers (optional)
export WHISPER_MODEL='base'        # tiny, base, small, medium, large
export WHISPER_DEVICE='cuda'       # cuda or cpu

# General
export LOG_LEVEL='INFO'            # DEBUG, INFO, WARNING, ERROR

For detailed provider configuration, see docs/PROVIDERS.md.

Supported Languages

en, es, fr, de, it, pt, auto (auto-detect)

Whisper supports 100+ languages.

Troubleshooting

Issue	Solution
Input file not found	Use absolute path: /full/path/to/video.mp4
API key not configured	export DEEPGRAM_API_KEY='...' or create .env
FFmpeg not found	Install: brew install ffmpeg (macOS)
TUI not working	uv sync --extra tui

For detailed troubleshooting, see docs/TROUBLESHOOTING.md.

Use Cases

Scenario	Input	Output	Time
Business meetings	2-hour recording	Executive summary, action items	~5-7 min
Training sessions	Multi-hour video	Searchable reference, key concepts	~10-15 min
Customer interviews	Interview recordings	Insights, pain points, feature requests	~3-5 min
Podcasts/Webinars	Long-form content	Chapter breakdown, topics, quotes	~5-10 min

Performance

• Accuracy: 95%+ (Deepgram Nova 3), 85%+ (Whisper large)
• Speed: Real-time (cloud), 0.5-5x real-time (local)
• Languages: 10+ (cloud), 100+ (Whisper)

Documentation

• Provider Configuration — Whisper, Parakeet, cloud provider setup
• Templates Guide — Customize Markdown output
• Troubleshooting — Common issues and solutions
• HTML Dashboard — Interactive dashboard rendering
• Examples — Sample outputs and scripts

Contributing

# Development setup
git clone https://github.com/edlsh/audio-extraction-analysis.git
cd audio-extraction-analysis
uv sync --dev --extra tui --extra parakeet

# Run tests
pytest                                    # Unit tests
./scripts/run_tests.sh --profile all      # Full suite

# Code quality
black src/ tests/ && ruff check src/      # Format + lint

License

This project is provided as-is for professional use.

---

Transform recordings into structured, actionable documentation.