Skip to content
/ AISL Public

Instruction-based audio storage - like source code for audio. 2-3x compression, built-in integrity, hyper-personalization.

License

Notifications You must be signed in to change notification settings

ISLHub/AISL

Repository files navigation

AISL - Audio Intelligent Storage Language

Part of the ISL (Intelligent Storage Language) family - Revolutionary instruction-based storage that enables compression, verification, and hyper-personalization

License: MIT Python 3.8+

⚠️ Important: This open source release covers ONLY the audio format implementation. Other ISL formats (video, image, text) and the underlying ISL architecture are subject to patent protection and remain proprietary. See LICENSE for full details.

What is AISL?

AISL (Audio Intelligent Storage Language) is the first open-source member of the ISL family - a paradigm shift from storing data to storing instructions for recreating data.

Instead of storing audio waveforms, AISL stores instructions for recreating audio. Think of it as "source code for audio" - editable, verifiable, and personalizable.

The ISL Vision

ISL (Intelligent Storage Language) represents a fundamental rethinking of data storage:

  • Traditional storage: Store the final output (waveform, pixels, text)
  • ISL approach: Store instructions for recreating the output

AISL proves this concept for audio. Future ISL formats will extend this paradigm to video, images, text, and more - but AISL is the first to be released as open source.

Key Features

✨ Instruction-Based - Stores HOW to create audio, not the audio itself
πŸ—œοΈ Compression - 2-3x compression with 0.99+ correlation
πŸ” Verified - Built-in cryptographic integrity checking
✏️ Editable - Modify audio by editing instructions (v2)
🎭 Personalizable - Insert names, adjust pitch/energy dynamically

Why Use AISL?

Feature MP3 FLAC Opus AISL
Instruction-based ❌ ❌ ❌ βœ…
Editable content ❌ ❌ ❌ βœ…
Built-in verification ❌ ❌ ❌ βœ…
Hyper-personalization ❌ ❌ ❌ βœ…
Good compression βœ… ❌ βœ… βœ…

Quick Start

# Install dependencies
pip install -r requirements.txt

# Encode audio to AISL
python aisl_encoder.py my_audio.wav

# Decode AISL back to audio  
python aisl_decoder.py my_audio.aisl output.wav

# Play AISL file directly
python aisl_player_gui.py my_audio.aisl

Demo

See the demos folder for:

  • Original vs encoded vs decoded files
  • Complete statistics and metrics
  • Quality comparisons

Results: 2.09x compression, 0.9937 correlation, 52% space saved

Use Cases

Storage & Distribution (v1)

  • πŸ“š Digital libraries & archives
  • πŸŽ™οΈ Podcast distribution
  • πŸ“– Audiobook platforms
  • 🎡 Music streaming
  • πŸ“‘ Bandwidth-limited apps

Personalization (v2)

  • 🎀 Insert listener names in podcasts
  • 🎢 Dynamic music adaptation
  • πŸ“• Interactive audiobooks
  • 🧘 Personalized meditation/fitness
  • 🌍 Localized content variations

How It Works

Encoding

  1. Extract voice anchors - Capture speaker characteristics
  2. Generate instructions - Timing, phonetics, pitch, energy
  3. Compute residuals - High-quality corrections
  4. Apply compression - Entropy coding
  5. Add integrity hash - Merkle tree verification

Decoding

  1. Verify integrity - Check Merkle tree
  2. Load anchors - Restore speaker characteristics
  3. Execute instructions - Generate base audio
  4. Apply residuals - Achieve high fidelity
  5. Output audio - Reconstructed waveform

Architecture

AISL File Structure:
β”œβ”€β”€ Metadata (sample rate, duration, etc.)
β”œβ”€β”€ Voice Anchors (speaker characteristics)
β”œβ”€β”€ Instructions (timing, phonetics, features)
β”œβ”€β”€ Residuals (quality corrections)
└── Integrity Hash (Merkle tree)

Python SDK

AISL provides a simple, clean SDK for integration:

from aisl_encoder import AISLEncoder
from aisl_decoder import AISLDecoder

# Encode
encoder = AISLEncoder()
encoder.encode_file("input.wav", "output.aisl")

# Decode
decoder = AISLDecoder()
audio, metadata = decoder.decode_file("output.aisl", "reconstructed.wav")

# Verify integrity
decoder.verify_integrity()  # Raises exception if tampered

Integration Examples

Streaming service:

# Convert library to AISL for storage savings
for track in music_library:
    encoder = AISLEncoder()
    encoder.encode_file(track.wav_path, track.aisl_path)

Content distribution:

# Verify files before playback
decoder = AISLDecoder()
try:
    decoder.decode_file("podcast.aisl", "temp.wav")
    decoder.verify_integrity()  # Ensures not tampered
    play_audio("temp.wav")
except IntegrityError:
    log_security_issue()

Hyper-Personalization (v2)

from v2.aisl_editor import AISLEditor

# Load AISL file
editor = AISLEditor()
editor.load_file("podcast.aisl2")

# Personalize!
editor.insert_name("Sarah", start_time=5.0)  # Insert listener name
editor.scale_energy(1.3)  # Make 30% louder
editor.shift_pitch(2)  # Up 2 semitones

# Save personalized version
editor.save("podcast_sarah.aisl2")

Installation

git clone https://github.com/ISLHub/AISL.git
cd AISL
pip install -r requirements.txt

Requirements

  • Python 3.8+
  • NumPy
  • SciPy
  • librosa
  • soundfile
  • PyTorch (for v2 features)

Documentation

Performance

Audio Original AISL v1 Ratio Quality
10s sample 960 KB 449 KB 2.09x 0.9937
Full song (145s) 26.5 MB 7.7 MB 3.44x 0.9824

Roadmap

  • Core encoding/decoding (v1)
  • Cryptographic verification
  • GUI player
  • Hyper-personalization framework (v2)
  • Real-time streaming support
  • Multi-language TTS integration
  • Mobile SDKs (iOS/Android)
  • Web player (WASM)
  • Plugin for DAWs

Contributing

We welcome contributions! See CONTRIBUTING.md for guidelines.

License

MIT License with important clarifications - see LICENSE for full details.

Key Points:

  • This license covers ONLY the AISL (audio) implementation
  • Other ISL formats (video, image, text) remain proprietary
  • Patent rights are reserved (see LICENSE for patent notice)
  • Trademark rights require attribution
  • Free for personal and commercial use of the audio format

Citation

If you use AISL in your research, please cite:

@software{aisl2024,
  title={AISL: Audio Intelligent Storage Language},
  author={Christopher Streicher},
  year={2024},
  url={https://github.com/ISLHub/AISL}
}

Support

Acknowledgments

Built on the concept of Instructional Storage Language - a paradigm shift from storing data to storing instructions for recreating data.


Note: AISL v2 hyper-personalization features (name insertion, voice synthesis) are in active development. Current implementation focuses on instruction framework and metadata tracking.

About

Instruction-based audio storage - like source code for audio. 2-3x compression, built-in integrity, hyper-personalization.

Resources

License

Contributing

Stars

Watchers

Forks

Packages

No packages published

Languages