It’s been some time since our last announcement about AcousticBrainz, so I wanted to give an update about where we are at and what our next steps are.
As of this week, we are stopping accepting submissions to AcousticBrainz and removing downloads of the submission tools.
If you want to continue computing features on your own audio files, you can still download the essentia feature extractor used in AcousticBrainz and run it yourself on your own audio files, or the updated version.
In the coming week we will be moving the AcousticBrainz website to a new server so that we can keep it running until at least the end of 2022. The API for downloading features from AcousticBrainz will remain working until we notify otherwise.
With the shut down of submissions, we will also publish a number of data dumps containing all data submissions. We have finished writing and testing the new code to do this, so this process will come soon, hopefully within the next week or so. These data dumps will contain json files for both lowlevel and highlevel data, for all submissions in AcousticBrainz (approximately 30 million items). Our current estimates with testing show that these dumps will be about 900-1000GB in size, compressed. Because of the size of these dumps we will also provide some additional dumps:
A dump of one file per MBID (taking into consideration MBID redirects). Approximately 7 million items
A dump of specific features (key, BPM), allowing you to download a much smaller dump if you only want some specific features.
If there are some other specific features that interest you, then let me know and we can include these in the dump
We will also make available a dump of the raw data used in the acoustic similarity endpoint, as well as the data files used for this endpoint so that you can perform acoustic similarity lookups without relying on the AcousticBrainz API.
Thanks for the work, sure there are many more people interested on continue working with that data.
Apart from BPM and Key, Mood data would be also interesting.
For ex.: Not acoustic; Not aggressive; Electronic; Not happy; Not party; Relaxed; Not sad
Also, how can we generate the mood , genre, … data? The AcousticBrainz extractor and the Essentia extractor binaries DON’T have Gaia pre-built so the svm_models can not be run.
There is not currently a binary with Gaia for windows available. Could you provide one? That way regular users could still make full use of the high level data when this shutdowns completely.
Hi,
we’ve released our data dumps, you can download the mood data from here.
However, you must currently download the full highlevel dump to obtain this data. I will see if it’s also possible to create a more simple data dump (perhaps a CSV) containing these seperate data values so that you can easily obtain just happy/not happy annotations, for example.
Regarding pre-compiled windows extractors with gaia support built-in, I will discuss with my colleagues at MTG and see if we can provide these. I will get back to you.
About the windows extractors, please, it would be really appreciated. Data it’s currently being used on this project as an offline music recommendation service to create playlists based on similarity:
I have been looking for weeks for binaries without luck. At the Essentia github no one has been able to help with that, there are dozens of issues there about cross-compiling failing and even another colleague has been trying to do it by himself on linux without luck. Essentia + gaia always fails and tutorials or dependencies must be out of date.
I will appreciate some guidance about cross compiling essentia tools with gaia for windows .
I can build the executables for Linux but when I tried to build the windows ones with gaia the build fails when trying to link with
/usr/bin/i686-w64-mingw32-ld: no se puede encontrar -lQtCore
/usr/bin/i686-w64-mingw32-ld: no se puede encontrar -ldl
collect2: error: ld returned 1 exit status
Without --with-gaia option the windows exes build.
I used the MTG/essentia github.
I followed the instructions without any luck, but like there is no explicit reference to building with gaia maybe I’m missing something.
Of course, downloadable executables will be good too.
Unfortunately, updates to a windows binary for a functionality that never existed in AcousticBrainz isn’t something that MetaBrainz is going to be able to help with. The AcousticBrainz site only worked with linux binaries of the high-level extractor, and the windows low-level extractor that we provided didn’t come with high level support.
There has been some work at the MTG to get gaia and essentia compiled with QT5, which means that it should be possible to get these binaries compiled again and working in windows, but this isn’t something that Metabrainz will be helping with.