Proposal: Capturing Vinyl

Every record played is captured to FLAC, by itself: put a record on and it is kept, track by track, and a track already kept is skipped. Nobody presses record. Over time the digital collection comes to include what is held only on vinyl, and every capture teaches EdS to name that record itself.

Why

• The vinyl-only records join the digital collection. A record with no CD or rip becomes playable anywhere the house plays music, and asked for by voice like any other.
• The turntable is named locally. EdS names a record today by sending a few seconds of it to Shazam, through an endpoint nobody promised (ADR-0021). A capture is fingerprinted, so the next time the record plays it is named from the owner’s own collection: the same pressing, the same mastering, even the same surface. Shazam is needed only for a record not yet heard.

What EdS already has

Capturing is a second use of what the turntable driver already hears, not new listening:

What it does not have is the sound. The driver hears the amplifier’s tape output in mono at 16 kHz, which is enough to name a track and not enough to keep one.

How it would work

1. A side begins, and the capture starts, at archival quality.
2. It is split into tracks by the gaps, held to the edition’s track list: how many tracks the side has and about how long each is, from Discogs or MusicBrainz. Gaps alone are fooled by a quiet passage or tracks that run into each other; the list alone does not know where the needle came down. Together they do.
3. Each whole track is kept as FLAC, tagged with what it is. A track cut short (the needle lifted, the side not finished) is thrown away, and kept on a later play.
4. A track already kept is skipped. EdS asks the inventory what is still missing from the edition; a record is fully captured when nothing is. Side A today and side B next month add up.
5. The capture is fingerprinted, so the next play is named locally.

The first play of a record still needs a name before there is a fingerprint: the recognizer names the track, as today, and the owner’s collection names the edition. When more than one pressing is owned, the original is taken, and can be corrected.

The boundary

EdS listens, captures and names; it knows the sound. The inventory knows music: a capture is to it one more copy of the vinyl edition, with its tracks, and it can say which are still missing. It hears nothing, as the music inventory proposal promises.

Open

• The capture’s sound. A card chosen for naming tracks is not one to archive with. The house already turns this stereo’s output into S/PDIF for its switch (Background); a digital input there would keep exactly what that converter makes, at its resolution and its quality. Otherwise, a converter chosen for the purpose. Which, and at what resolution.
• Where the files go. The media server is the natural home. Written there with tags naming the MusicBrainz IDs and the inventory’s edition, they would be collected with MusicBee’s library, already identified. At 24-bit/96 kHz an LP is on the order of a gigabyte, so copies are rate-capped (INC-0002).
• Where it runs. On the agent Pi beside the driver, or a device of its own on the same sound.
• Splitting a side whose tracks run into one another, and a record whose track list is not known.
• What is kept of the surface: clicks and pops left as played, or a copy cleaned up.
• Vinyl FLAC or CD FLAC, for an album held both ways: which the house plays first is the owner’s preference, not a rule (eds-inventory’s “best copy”).