Invisible Audio Watermarking for Leak Attribution & Theft Protection
Serialize unreleased masters, beats, stems, and screeners with imperceptible acoustic signatures. Trace leaks back to the exact recipient—even through lossy MP3 compression, social media re-encodes, and phone speaker recordings.
EVIDENCE & COMPLIANCE
Watermark Detection Audit Reports
Aureal reports whether a watermark ID was detected and lets you compare it with distribution records. A match is one data point, not independent proof of authorship, ownership, or chain of custody.
File Hashes and Audit Metrics
A report can record SHA-256/SHA-512 hashes and detection metrics for an analyzed file. Preserve the original files, timestamps, and independent custody records; the report alone does not establish chain of custody or legal admissibility.
Detection Metrics
Reports can include signal-detection measurements such as confidence, bit error rate, and CRC status. Interpret these metrics in the context of the source audio and processing.
Report Integrity Seal
Reports can include a shared-secret HMAC integrity seal. It is not a public-key signature and does not independently verify who created the report or when.
01
Seamless Studio Ingestion
Embed recipient-specific watermark IDs into master recordings, stems, and screeners; audibility and detection depend on the material and settings.
02
Distribution & Attack Resilience
Detection has been tested against selected codecs and acoustic transformations; results vary with the source audio, codec, and processing.
03
Deterministic Attribution
Scan a suspected leak for a matching watermark ID, compare the result with distribution records, and export a detection audit report.
ENTERPRISE WORKFLOW INTEGRATION
1-Command Deployment & REST Microservice
Deploy as an air-gapped Docker container, invoke from your Python/Go pipelines, or use the zero-dependency CLI.
# 1. Start the air-gapped microservice (zero cloud calls, port 8080)
docker compose -f docker/docker-compose.yml up -d
# 2. Embed watermark via binary audio stream (WAV, MP3, FLAC)
curl -X POST "http://localhost:8080/v1/embed?id=928401&format=mp3" \
-H "Content-Type: audio/wav" \
--data-binary @master.wav \
-o watermarked.mp3
# 3. Detect and generate audit report
curl -X POST "http://localhost:8080/v1/detect?id=928401&report=true" \
-H "Content-Type: audio/mpeg" \
--data-binary @leaked_clip.mp3# Direct MP3, FLAC, and WAV embedding with automatic FFmpeg fallback
auralwatermark embed master.wav protected.mp3 --id 928401
# Scan and export forensic audit report and text certificate
auralwatermark detect leak.mp3 --id 928401 --report proof.json --report-txt cert.txt
# Launch standalone offline GUI studio
auralwatermark gui// Out-of-the-box autocomplete via types/index.d.ts
import { embedWatermark, detectWatermark, generateForensicReport } from "aureal-watermark";
import { readWavFile, writeWavFile } from "aureal-watermark/src/wav.js";
const wav = await readWavFile("master.wav");
const protectedAudio = embedWatermark(wav.samples, wav, { payloadId: 928401 });
await writeWavFile("protected.wav", protectedAudio, { ...wav, bitDepth: 16 });
const res = detectWatermark(protectedAudio, wav, { payloadId: 928401 });
console.log(res.detected, res.confidence); // true, 1.0
ENTERPRISE CONTENT PROTECTION
Engineered for Major Studios, Record Labels & Rights Holders
How acoustic watermarks can complement metadata and fingerprinting in audio provenance workflows; retain independent evidence for chain of custody.
Pre-Release Screener & Promo Leak Attribution
When distributing unreleased music, film cues, stems, or review screeners to 200 journalists, radio stations, and mixing engineers, serialize every copy with a unique 32-bit recipient ID. If a track leaks to Telegram, Discord, or torrent trackers, recover the exact recipient within 350ms.
Lossy Transcoding & Social Media Resilience
Pirated leaks are never uncompressed WAV. Pirates compress to 64–128 kbps MP3, upload to TikTok / YouTube, or record audio through phone speakers. Aureal's DSSS spread-spectrum modulation embeds beneath biological psychoacoustic curves, surviving extreme lossy re-encoding and pitch shifts.
AI Voice Model & Training Set Ingestion Audit
Forensic watermarks can help you audit distributed audio. Validate results against your own files and keep the original master and distribution record as evidence.
Local Audio Processing
Aureal's audio embed and detection operations run locally. Optional commercial-license activation contacts Polar.sh; install and distribute the required assets according to your security requirements.
Human Auditory System (HAS) Masking Curves
Metadata can be stripped during export or re-encoding. Aureal explores spread-spectrum embedding under a perceptual threshold; test the result with your own delivery pipeline.
Frequency Masking Thresholds
When a strong musical tone plays (e.g. 1 kHz snare transient), adjacent frequency bands are biologically rendered inaudible to the human cochlea:
// Computes Bark scale psychoacoustic mask
const maskLevel = Psychoacoustics.calculateThreshold(fftBuffer);
const spreadSignal = DSSS.modulate(watermarkPayload, pseudoRandomSequence);
// Scales watermark energy strictly below inaudibility envelope
const output = fftBuffer.injectSubThreshold(spreadSignal, maskLevel);
Direct Sequence Spread Spectrum (DSSS)
The watermark payload is spread across thousands of frequency bins using an orthogonal cryptographic key sequence.
Because the signal energy in any single bin is orders of magnitude below ambient acoustic noise, lossy audio codecs cannot compress or filter it away without destroying the underlying music.
const detectedId = AurealScanner.correlate(audioStream, clientKey);
assert.equal(detectedId.valid, true);
// Recovers 64-bit cryptographic ownership signature
Audio Attribution Solutions Compared
Why ID3 metadata tags and acoustic fingerprinting fail to stop AI scraping and music theft.
| Watermarking Technique | ID3 Metadata Tags (MP3 / FLAC) | Acoustic Fingerprinting (Shazam) | Aureal Acoustic Spread-Spectrum |
|---|---|---|---|
| Lossy Compression Resistance | Stripped immediately on YouTube/TikTok re-encode | Database matching only | Tested in selected scenarios transcoding |
| Individual Recipient Tagging | Impossible (File hash is shared) | Cannot distinguish two identical tracks | Unique Invisible Serial ID per preview recipient |
| AI Voice Model Scrape Audit | Ignored by web scrapers | Fails on spliced synthetic voice training | Verifiable Provenance Signature inside training data |
| Offline Private Scanning | Trivial text inspection | Requires central corporate database | Client-Side Demo Scanner in browser |