Disc ID Calculation

From MusicBrainz Wiki
Jump to navigationJump to search
cdhand.gif

Let us first have some words about how data is organized on a CD, going top down.

An Audio CD (CD-DA) can hold up to 99 audio tracks.

Sampling is done at a rate of 44.1 kHz using 16 bits resolution per channel, thus there are 44100 x 2 bytes x 2 channels (Stereo) = 176400 bytes of PCM data stored per second.

This audio data is contained in logical blocks of 2352 bytes each on the CD, holding 2352 / 176400 = 1 / 75 seconds of sound.

A logical block plus 882 bytes of error correction and control data forms a raw block of 3234 bytes that is spread among 98 frames of 33 bytes each that are all together written on one spiral track among the CD.

At present we can stop at the level of logical blocks for our task. The deeper stuff will be needed in case the CD-Text format becomes popular, which stores extra data (like artist and track information) in the above mentioned control data parts.

Reading the TOC

Audio CD

Now let us have a look at a real world CD-DA. Using a tool like cdrecord on this CD will yield:

$ cdrecord dev=/dev/cdrom -toc
...
first: 1 last 6
track:   1 lba:         0 (        0) 00:02:00 adr: 1 control: 2 mode: 0
track:   2 lba:     15213 (    60852) 03:24:63 adr: 1 control: 2 mode: 0
track:   3 lba:     32164 (   128656) 07:10:64 adr: 1 control: 2 mode: 0
track:   4 lba:     46442 (   185768) 10:21:17 adr: 1 control: 2 mode: 0
track:   5 lba:     63264 (   253056) 14:05:39 adr: 1 control: 2 mode: 0
track:   6 lba:     80339 (   321356) 17:53:14 adr: 1 control: 2 mode: 0
track:lout lba:     95312 (   381248) 21:12:62 adr: 1 control: 2 mode: -1

You can see that the CD has 6 audio tracks and the special lead-out track (it has number 170 in the CD TOC). Also note that the LBA (Logical Block Address) offsets start at address 0, but the first track starts actually at 00:02:00 (the standard length of the lead-in track). So we need to add 150 logical blocks to each LBA offset. The resulting data needed to calculate a MusicBrainz disc ID are:

track 1:       150    (150 + 0)
track 2:       15363  (150 + 15213)
track 3:       32314  (150 + 32164)
track 4:       46592  (150 + 46442)
track 5:       63414  (150 + 63264)
track 6:       80489  (150 + 80339)
lead-out track: 95462  (150 + 95312)

Multi-session (audio + data) CD

Because MusicBrainz doesn't include data tracks in Disc IDs, reading the TOC from a multi-session disc is a little more complicated. Running cdrecord on this CD will give us:

$ cdrecord dev=/dev/cdrom -toc
...
first: 1 last 8
track:   1 lba:         0 (        0) 00:02:00 adr: 1 control: 0 mode: 0
track:   2 lba:     13959 (    55836) 03:08:09 adr: 1 control: 0 mode: 0
track:   3 lba:     33436 (   133744) 07:27:61 adr: 1 control: 0 mode: 0
track:   4 lba:     52927 (   211708) 11:47:52 adr: 1 control: 0 mode: 0
track:   5 lba:     65631 (   262524) 14:37:06 adr: 1 control: 0 mode: 0
track:   6 lba:     77742 (   310968) 17:18:42 adr: 1 control: 0 mode: 0
track:   7 lba:     99024 (   396096) 22:02:24 adr: 1 control: 0 mode: 0
track:   8 lba:    125824 (   503296) 27:59:49 adr: 1 control: 6 mode: 1
track:lout lba:    188333 (   753332) 41:53:08 adr: 1 control: 6 mode: -1

This is an example of a CD with an extra track of data (what you know as CD-ROM), marketed in this case as CD-Extra, featuring a video and some pictures. (More precise: this is disc with two sessions, audio and data)

This CD has 8 tracks, but only the first 7 audio tracks should be used to calculate a MusicBrainz disc ID. The problem is that we can't use the offset of track 8 as the "lead-out track" offset, because there is a gap between the audio session and the data session. This gap is 11400 frames long (11250 frames for lead-out/lead-in + 150 frames of pre-gap), so we need to substract this value from the offset of track 8 to get the end of track 7. The result is:

track 1:       150    (150 + 0)
track 2:       14109  (150 + 13959)
track 3:       33586  (150 + 33436)
track 4:       53077  (150 + 52927)
track 5:       65781  (150 + 65631)
track 6:       77892  (150 + 77742)
track 7:       99174  (150 + 99024)
lead-out track: 114574 (150 + 125824 - 11400)

Calculating the Disc ID

Step 1: Hashing the binary TOC data

The CD Index algorithm simply takes the following pieces of data and runs them through the SHA-1 hash function:

  • First track number (normally one): 1 byte
  • Last track number: 1 byte
  • Lead-out track offset: 4 bytes
  • 99 frame offsets: 4 bytes for each track
  • If there are less than 99 tracks (almost certainly), the value 0 will be used instead.

Before the data is fed through the SHA-1 hash, it is converted to upper-case hex ASCII using printf("%02X", value); for single-byte values and printf("%08X", value); for 4-byte values.

Code is a better definition than English, so here is the code that calculates the DiscID:

sprintf(temp, "%02X", pCDInfo->First­Track);
sha_update(&sha, (unsigned char*) temp, strlen(temp));

sprintf(temp, "%02X", pCDInfo->Last­Track);
sha_update(&sha, (unsigned char*) temp, strlen(temp));

for (i = 0; i < 100; i++) {
    sprintf(temp, "%08X", pCDInfo->Frame­Offset[i]);
    sha_update(&sha, (unsigned char*) temp, strlen(temp));
}
sha_final(digest, &sha);

Note that the lead-out track is stored in pCDInfo->Frame­Offset[0].

Step 2: Base64-encoding of the hash

The resulting 20 byte SHA-1 signature is converted into a Base64 encoded string of printable ASCII characters, which is the disc ID. All disc ID strings are thus exactly 28 characters long. In the above audio CD example the disc ID is 49HHV7Eb8UKF3aQiNmu1GR8vKTY-. For details about Base64, please see RFC 4648 or the reasonable good Wikipedia article.

Important note: The Base64 encoding used by MusicBrainz is not the same one as specified in RFC 4648. The specification uses +, /, and = characters, all of which are special HTTP/URL characters. To avoid the problems with dealing with that, MusicBrainz uses ., _, and - instead. For details on this, please refer to base64.c in the libdiscid source code.

Remarks

The disc ID scheme has the advantage of being very simple (simple to understand, simple to implement). However, two different pressings of the same disc may have different IDs. To handle this case, the MusicBrainz system will let a user check to see if the CD already exists in the system under a different ID. If so, the system creates a new association for the different pressing of the same CD. Also note that a disc ID is usually not ambiguous, but it can still happen that different CDs have exactly the same set of frame offsets and hence the same disc ID, for example lwHl8fGzJyLXQR33ug60E8jhf4k-.

If you'd like to know more about disc ID calculation, please download the libdiscid source code and check it out. The code is clean and self documenting.

If you are interested in creating other MusicBrainz clients and need the SHA-1 source code, check out RFC 3174.

Tools

Tools based on our library libdiscid:

  • Our tagger Picard that has a feature “Lookup by CD”
  • Some of the ISRC submit tools that also support submitting Disc ID

Tools based on third party implementations:

  • mbdiscid, a command-line tool (written in Perl) that computes the Disc ID, and either prints it or submits it through a browser
  • MetaBrainz.MusicBrainz.dotnet-mbdiscid (NuGet Package), a command-line tool (using the C#/Mono/.NET library MetaBrainz.MusicBrainz.DiscId) that shows information about an audio CD, including the Disc ID, CD-TEXT data, ISRC values…

Libraries

It can be calculated using our C library, libdiscid, or any of its many language bindings.

Third party libraries:

  • C#/Mono/.NET - MetaBrainz.MusicBrainz.DiscId (NuGet Package), a library which adds support for CD-TEXT retrieval, but does not currently support Solaris (no recent Mono, no .NET Core) or macOS (no machine available)

Links

  • The How Stuff Works site has a very readable introduction on CD technology intended for the technically curious. Written by Marshall Brain.
  • There is actually a very fine FAQ about toasting available that has lots of interesting facts. Written by Andy McFadden.