01 Ingest
Built for files of any size
Audio streams in as 64 KB blocks, so memory stays flat whether the file is three minutes or three hours.
stream64 KB blocks
peak memory~12 MB · flat
AI Audio Forensics · 100% In-Browser Every WAV file,
Every WAV file,
proven.
Detect AI-generated audio, inspect every metadata chunk, and strip what you don't want — free, private, and entirely inside your browser.
scanning signal…
Fingerprints generators & credentials from
Suno Udio ElevenLabs OpenAI C2PA
Live detection
Watch it weigh the evidence
Load a file and ProvenanceX scans the bytes — content credentials, generator fingerprints, and the waveform itself — then returns a verdict and the exact signals behind it.
- A probability score in well under a second
- Every signal shown — not just a yes / no
- Runs locally; nothing is uploaded
sample-track.wav 48 kHz · 24-bit · 3:52
0
Verdict
◆Likely AI-generated
A C2PA content credential and a generator fingerprint were found in the file's metadata.
Signals found3
C2PA content credential
Generator fingerprint
Acoustic heuristic
0:003:52
Chunks · keep / stripstrip saves 1.24 MB
fmtKEEP dataKEEP factKEEP LISTSTRIP C2PASTRIP bextSTRIP iXMLSTRIP ID3 STRIP
The pipeline
One pass through the instrument
A file flows through every stage in a single pass — detected, read, and cleaned — and never leaves your device.
02 Detect
AI origin scan
Reads C2PA credentials, generator fingerprints from Suno, Udio, OpenAI and ElevenLabs, and acoustic heuristics — returning a probability and the exact signals found.
+c2pa.manifestissuer
cai.suno+LIST/INFO·ISFTmatches
udio~spectral.profilereads
synthetic verdict AI-GENERATED p = 0.00
03 Inspect
Byte-level metadata
Every chunk laid out with its size and keep/strip status — fmt, data, LIST, bext, iXML, ID3, C2PA.
fmt16 B
data41.6 MB
LIST/INFO1.24 KB
04 Strip
Remove what doesn't belong
Drop non-essential chunks while keeping fmt and data, so the file stays byte-clean and perfectly playable.
stripped6 chunks
saved−1.24 MB
✓ Output
A clean, verified WAV
Out the other side: the same audio, stripped and documented — written entirely in your browser.
outputsample-track.clean.wav
size40.4 MB
status✓ playable
End-to-end on your device. 0 bytes uploaded · 0 network requests · runs on WebAssembly.
The inspector
Inspect every byte
A precise readout of the file's structure — audio properties on the left, every chunk and its verdict on the right.
Chunk map · 8 detected 92% · Likely AI
| Chunk | Type | Size | Status |
|---|---|---|---|
| fmt | format | 16 B | KEEP |
| data | audio samples | 40.31 MB | KEEP |
| fact | sample count | 4 B | KEEP |
| LIST | INFO tags | 0.74 MB | STRIP |
| C2PA | content credential | 0.47 MB | STRIP |
| bext | broadcast WAV | 858 B | STRIP |
| iXML | metadata | 4.2 KB | STRIP |
| ID3 | tags | 12.4 KB | STRIP |
Strip 5 chunks · keep playback intact · save 1.24 MB Download clean WAV ↓
Under the hood
It reads the whole file
A WAV is a RIFF container — a header and a chain of chunks. ProvenanceX walks every one, byte for byte, to find what's there and what to remove.
RIFF / WAVE structure sample-track.wav · 41.6 MB
Essential chunks are kept so the file stays playable; everything non-essential is flagged to strip.
| Offset | Chunk | Size | Status |
|---|---|---|---|
| 0x000000 | RIFF·WAVE | CONTAINER | |
| 0x00000C | fmt | 16 B | KEEP |
| 0x000024 | fact | 4 B | KEEP |
| 0x000030 | ds64 | 28 B | KEEP |
| 0x000050 | data | 41.6 MB | KEEP |
| 0x2999CF8 | LIST/INFO | 842 B | STRIP |
| 0x299A042 | bext | 601 B | STRIP |
| 0x299A29B | iXML | 1.10 KB | STRIP |
| 0x299A6F4 | ID3 | 312 B | STRIP |
| 0x299A82C | C2PA | 18.4 KB | STRIP |
Chunks padded to even bytes — output stays compliant with every modern & legacy DAW.
Where the signatures live 5 sources
Known credential issuers and generator fingerprints, matched against the chunks above.
S credential
Suno
in
C2PA + LIST/INFO U string
Udio
in
LIST/INFO · ISFT string 11 credential
ElevenLabs
in
C2PA · encoder signature AI credential
OpenAI
in
C2PA content manifest ~ signal
Acoustic heuristics
in
data · spectral profile Built like an instrument
Precise, safe, and fast
MEM · 01
O(1)
Memory footprint
Constant memory, even on multi-GB files.
I/O · 02
0 KB
Streaming blocks
Audio streamed in fixed-size chunks.
NET · 03
0 B
Bytes uploaded
Everything runs locally via WASM.
FMT · 04
0%
DAW-compliant
Even-byte chunk alignment preserved.
How it works
Three steps, zero uploads
Load a WAV
Drag & drop a file onto the app or pick one. It loads instantly — no upload, no account.
Inspect & detect
View audio properties and every chunk, then run the AI origin diagnostic for a verdict and the signals behind it.
Strip & download
Remove non-essential metadata and download a clean, fully-playable WAV in one click.
FAQ
Frequently asked questions
Commands
$provx
Start now
Launch ProvenanceX →