An open-source, public domain audio format that natively supports multi-stem architecture, neural compression, smart metadata, spatial audio, P2P streaming, and non-destructive editing.
One file to rule them all — production, distribution, playback, and archival.
Specification • Getting Started • Architecture • Roadmap • Contributing
Current audio formats were designed in the 1990s for a simpler world. They store flat, single-stream audio with basic metadata. NEO reimagines audio from the ground up:
| Capability | MP3/AAC | FLAC/WAV | NI Stems | NEO |
|---|---|---|---|---|
| Multi-stem (vocals, drums, bass...) | ❌ | ❌ | ✅ (4 fixed) | ✅ (1-8 dynamic) |
| AI neural compression (90x ratio) | ❌ | ❌ | ❌ | ✅ |
| Bit-perfect lossless mode | ❌ | ✅ | ❌ | ✅ (hybrid) |
| 3D spatial audio (object-based) | ❌ | ❌ | ❌ | ✅ |
| Provenance / anti-deepfake (C2PA) | ❌ | ❌ | ❌ | ✅ |
| Smart metadata (JSON-LD) | ID3 tags | Vorbis | MP4 | ✅ (linked data) |
| Temporal lyrics + chords + BPM | ❌ | ❌ | ❌ | ✅ |
| Web3 royalties / direct payment | ❌ | ❌ | ❌ | ✅ |
| P2P / IPFS native streaming | ❌ | ❌ | ❌ | ✅ |
| Non-destructive edit history | ❌ | ❌ | ❌ | ✅ (git-like) |
| Progressive loading (instant play) | ❌ | ❌ | ❌ | ✅ |
| Open source / patent-free | Expired | ✅ | ✅ | ✅ (Unlicense) |
Every .neo file can contain up to 8 separate audio stems (vocals, drums, bass, melody, etc.). Mute the vocals for instant karaoke. Isolate the drums for sampling. Or listen to the full mix — all from one file.
Powered by the Descript Audio Codec, NEO achieves ~90x compression at perceptual transparency. Combined with a FLAC residual layer, you get lossy streaming AND lossless archival in the same file.
Audio stems are positioned in 3D space as objects. Your player automatically renders them for headphones (binaural/HRTF), surround systems (5.1/7.1.4), or stereo — no re-mastering needed.
Cryptographic signatures prove whether a track is human-made, AI-generated, or a hybrid. Fight deepfakes and verify authenticity at the format level.
Start playing instantly with the embedded ultra-light preview. Higher-quality layers download in the background. Zero buffering, even on 4G.
Cut the intro? Boost the bass? Your edits are stored as metadata — the original audio is never touched. Revert to any version, like git for audio.
NEO Project
├── crates/ # Rust workspace (core implementation)
│ ├── neo-format/ # Container format read/write
│ ├── neo-codec/ # Audio codecs (DAC, FLAC, Opus)
│ ├── neo-metadata/ # JSON-LD, C2PA, Web3, temporal
│ ├── neo-spatial/ # 3D audio, Ambisonics, HRTF
│ ├── neo-stream/ # Merkle trees, P2P, progressive loading
│ ├── neo-edit/ # Non-destructive editing engine
│ ├── neo-cli/ # Command-line tool (`neo`)
│ └── neo-ffi/ # C bindings for cross-language interop
├── neural/ # Python ML components
│ └── neo_neural/ # DAC fork, ONNX export, source separation
├── spec/ # Format specification documents
└── tests/ # Unit, integration, and e2e tests
| Component | Technology |
|---|---|
| Core language | Rust (memory-safe, WASM-compilable) |
| Neural codec | DAC (Descript Audio Codec) fork, ONNX inference |
| Lossless layer | FLAC |
| Preview layer | Opus |
| Metadata | JSON-LD (schema.org) |
| Provenance | C2PA (c2pa-rs) |
| Hashing | BLAKE3 |
| Content addressing | Multihash / CID (IPFS-compatible) |
| Spatial rendering | Ambisonics + HRTF |
| ML framework | PyTorch → ONNX → ort (Rust) |
- Rust 1.75+
- Python 3.10+ (for neural codec components)
# Clone the repository
git clone https://github.com/CICCADA-CORP/NEO.git
cd neo-codec
# Build the Rust workspace
cd crates
cargo build --release
# The CLI tool is at target/release/neo
./target/release/neo --help# Encode a WAV file with 2 stems
neo encode vocals.wav instrumental.wav -o track.neo --stems vocals,instrumental
# Decode back to WAV
neo decode track.neo -o output/ --all-stems
# Inspect file info
neo info track.neo
# Play with stem control
neo play track.neo --mute vocals # Instant karaoke!
neo play track.neo --solo drums # Isolate drums
# Tag with metadata
neo tag track.neo --title "My Song" --artist "My Band" --bpm 120 --key Cm
# Download AI models
neo model download 44khzThis project is released into the public domain under the Unlicense.
You are free to use, modify, distribute, and build upon this work for any purpose, without any restrictions. No attribution required.
We welcome contributions of all kinds! Please read our Contributing Guide to get started.
- Format specification review — Help refine the binary format
- Rust implementation — Core container, codecs, and CLI
- Neural codec optimization — DAC fine-tuning for stems
- Testing — Audio quality metrics, edge cases, fuzzing
- Documentation — Tutorials, API docs, examples
NEO is not just a format — it's the PDF of audio.
Universal. Editable. Interactive. Secure.