Hacker News Reader

Comment by margorczynski | Hacker News Reader

I think it operates on the absolute state of the audio where e.g. Reaper uses deltas - your current result is simply all of the deltas applied on after the other so it is easy to go back with each change by just removing a delta from the sequence. Quite similar to e.g. Event Sourcing vs your regular source of truth in the DB

View on HN

I don't think this is the right description.

DAWs (digital audio workstations) edit by using metadata. Suppose you have an audio file that is 2 minutes long. You start using it in the application. The application notes "we have 2 minutes of audio, taken from the start of this file". Now if you "delete" the last minute of the file, nothing is done to the file, but the "metadata" inside the DAW now says "we have 1 minute of audio, taken from the start of this file".

If you then copy 20 seconds from the middle of that section, the DAW refers to this with metadata essentially saying "20 seconds of audio starting 20 seconds into this file".

Nothing is ever done to change the contents of the actual file.

ps. I write a DAW for a living.

The descriptions seem equivalent, to me. I tend to think of it as: a "source" is either raw audio, or the result of applying a parameterised filter to zero or more sources.

The reason it's not equivalent to me is that the concept of "a series of deltas" does not make clear that what is edited (with or without a delta model) is metadata. No DAW ever edits an audio file (except for a few specific operations where the user explicitly requests that). All the edits are carried out on metadata (some would call them an EDL - edit decision list; some would use other names). But the delta model is orthogonal to this.

> No DAW ever edits an audio file (except for a few specific operations where the user explicitly requests that).

This feels like semantics. No self-respecting DAW will overwrite your source files, but the DAW's project file is, technically, an audio file. And here's a 30-second video file:

  A = BlankClip()
  B = A.Subtitle("Hello, world!")
  Dissolve(A, B, 30)

(Modified from: https://en.wikipedia.org/w/index.php?title=AviSynth&oldid=11...)

You're using "audio file" in a domain-specific way. The common meaning of "audio file" is "a file that contains audio": a .aup file is an audio file, just the same as a GarageBand project is, to a room full of 14-year-old amateur musicians.

when i was a 14 (or even 12) year old amateur musician it was very obvious to me that DAW projects weren't audio files, I have never ever seen anyone saying this except your post across all my childhood friends who were dabbling with "artisanal" versions of FL studio, cubase SX, etc etc

The people I'm referring to were using GarageBand, MIDI, wav, mp3, and Musescore in the same kinds of sentences, and talked about converting between them. They also talked about importing into Sibelius and GarageBand. "Convert it [Musescore] to Sibelius" feels like it was about as common as "import the [Musescore] file into Sibelius". (And no, they hadn't heard of Lilypond.) Maybe I misunderstood what they meant. They were predominantly "play on an actual physical instrument"-type people, not tracker enthusiasts: without adult influence, perhaps two people present would have been using computer software at all. (Manuscript paper for everyone, except the one kid who only knew tab.)

Regardless, seems I'm the one with the uncommon use of language, here.

I think the description of metadata is good (my mental model is that a DAW operates on a stacked-up pile of operations to apply on top of the sources, and editing is all just adding/removing/modifying the parameters of those operations, which I think is equivalent to what you're saying), but the simple "doesn't change the contents of the actual file" standard I'm not sure is so useful. It would sweep up old Audacity and something like GIMP into the "nondestructive" pile just by virtue of each operating on a model where pre-existing source files get imported into the tool's native format and have to be exported back out.

Edit a WAV or JPEG with either of those and they'll dutifully not touch the source file (unless/until you tell them to export on top of it) despite internally operating directly on the data. (I think both have moved and are moving to offer more nondestructive modes of operation, but speaking of their legacy setups, anyway).

Interesting. It's like a piece table[0] but for audio.

[0] https://en.wikipedia.org/wiki/Piece_table

Reply on news.ycombinator.com