Analyzing music files using AI

DEVONthink
1

I have a number of mp3 music files that I have indexed which have unknown titles and are performed by unknown bands. I would like to be able to pass these to chatgpt4 and use it to determine the title of the song, its band and maybe other metadata that may be available. I cannot figure out how to do this from within DEVONthink 4. even better, since this is highly repetitive, I would like to be able to accomplish this through a smart rule. I have a paid account and api key so that’s not a problem. any suggestions?

2
  • Why would you want to do this in DEVONthink?
  • Have you actually tried this via ChatGPT online?
2 Likes
3

First you would have to transcribe the speech. If that’s successful, then the Chat - Query action of batch processing and smart rules might be able to return the desired metadata.

4

@Ghausler I’m also wondering why you reach for DEVONthink here.

There already exists a specialized tool for this: MusicBrainz Picard. It can scan a file to generate an acoustic fingerprint, which it then matches against a large database to identify the song and fetch metadata. The database doesn’t have every recording ever produced, but it has a lot. I doubt any AI will do better or even come close.

I haven’t tried automating it, but check the documentation. Probably see Command and Batch Processing + Appendix C: Command Line options. For details on fingerprinting, see Understanding Acoustic Fingerprinting and AcoustIDs

If you’re particular about your metadata, Picard also has many plugins to customize various aspects or add extra functionality. Easily installed directly from the settings.

There are also other tools that make use of the MusicBrainz database. I’ve been curious to try Beets, which looks like a strong contender for the best overall option to manage a music library + metadata, as long as you’re fine using the terminal.

(Edit: Replied to cgrunenberg instead of the topic by accident. Discourse won’t allow me to delete and post again because the content is “too similar”.)

1 Like
5

Picard…That’s a much better tool…purpose fit for this situation. Thank you.

6

Yes, it works but not particularly well. But are there other use cases where one would want to use Devonthink to manage media files and perhaps use AI to augment Devonthink’s native capabilities?

7

Thank you. That sounds promising.