I have spent the last couple of weeks processing and uploading AcousticBrainz results. At the present rate I may have 3-4 weeks left. I have a 4-cpu I7 system and it is brought to its knees with this processing. Looking at the performance monitors it does not use a GPU, but that may be because I have just the embedded GPU.
Is there (or will there be) an effort to use GPU’s for this processing?
Reading through the discussion here I see that the design is to multi-thread using N-1 CPU’s, which explains the CPU usage I am seeing. Is there a way of limiting this to N-2 without recompiling the code?
In specific response to this question, it’s actually a question about if essentia can use a GPU to compute audio features (as the AcousticBrainz extractor is actually one of the audio feature extractors provided by essentia). Essentia doesn’t have the functionality to use a GPU to compute this kind of audio features. I don’t think they will be planning on adding this support in the future.
Unrelated to the specific type of data extraction performed in AcousticBrainz, essentia does have some support for using deep learning models to predict some audio characteristics and this can use a GPU, however we don’t have current plans to integrate this into any AB tools.