Provir
Independent provenance for recorded music. We establish whether a file is what it was sold as.
The problem
A great deal of music sold as lossless is not. It was compressed once, decompressed and repackaged, and the buyer is charged for quality that is no longer in the file.
The industry knows. Distributors tell uploaders, in their own published guidelines, not to submit lossy files converted to FLAC. What nobody ships is a way to check. Every available tool looks for a missing top end, and above about 256 kbps there is very little left to find. The best of them still separates the two classes at that rate — but only at a false-positive rate far too high to accuse a named release, and none of them tells you which encoder was used.
The evidence
Provir reads what the encoder left behind in the audio itself. It needs no original to compare against and no metadata to trust.
| MP3 instrument | |
|---|---|
| MP3 320 kbps, full-band sources | 23 / 24 identified |
| MP3 320 kbps, band-limited sources | 0 / 12 identified |
| MP3 V0, full-band sources | 24 / 24 identified |
| MP3 V0, band-limited sources | 5 / 12 identified |
| genuine masters | 0 / 36 false positives |
| five other codecs | 0 / 180 misread as MP3 |
| Vorbis instrument | |
| Ogg Vorbis, quality 8 | 26 / 36 identified |
| genuine masters | 1 / 36 false positives |
| six other codecs | 0 / 216 misread as Vorbis |
Each set held 36 sources: 24 full-band and 12 band-limited, filtered before encoding. The filter removes most of what the MP3 instrument needs at these rates, so we show both halves rather than only the one we win: over all 36 sources that is 23 at 320 kbps and 29 at V0.
MP3 and Vorbis are read by two different mechanisms, so neither result depends on the other being right. The other-codec rows are about attribution: a zero means the instrument never mistook another codec for its own, not that it can detect those codecs.
The Vorbis instrument produced one false positive, on a lossless file that had been band-limited. We have since refined its fire rule, but the second set we checked the refinement on turned out not to settle the question, so no refined figure is claimed here. We report the result the sealed instrument actually gave.
Never wrongly accusing a genuine master matters more to us than catching every fake. The blind counts above are per file, as the evaluator scored them; when we price false positives on our own collection we count releases rather than files, the stricter of the two conventions.
Who it is for
Listeners and artists, who should be able to check what they bought or what they made without paying for the privilege. And anyone who sells recorded music and would rather find this in their own catalogue than have someone else find it for them.
Status
In active development and not yet publicly available. If you would like to be told when it is, or you audit a catalogue for a living, write to contact@provir.org.