Very good point. This is an active area of research, there's been some work in using other similarity measures like MI,Kendall's Tau etc in embeddings as well. And they have been found to perform pretty well.
EDIT: As @rubatuga said,they have been avoided largely as embeddings are continuous random vars and MI is for distributions. Nonetheless I think there can be merit in exploring these
PS: I am the author and just wanted to talk about the familiar basic stuff on this one
How about a vector oriented 'database' instead? Pinecone(https://www.pinecone.io/) does both exact and approx search and it's fully managed so you don't have to worry about reliability,availability etc.
That is definitely an interesting use case. You can look for methods that combine multiple vectors to create a combined vector or you may want to filter on metadata and vectors both. Who knows maybe you might be able to come with something very unique for your example!
Any search domain is good for vector search if your method for producing vectors is good! Seems like it should be possible with Recipes.Are you performing search based on recipe text? It shouldn't take long to try out a simple POC with standard text embedding models like (Glove,BERT etc) and observe vectors via tSNE or just using a vector index.
Looking at the VGGish paper itself, I see they use spectrograms as inputs, they show results where they can identify instrument types. I'm not too sure how specific embeddings from these models can be. Do we know if spectrograms can differentiate between two people's voice?
I found something for MI if you wanna check it out : https://www.aclweb.org/anthology/2020.acl-main.741.pdf
EDIT: As @rubatuga said,they have been avoided largely as embeddings are continuous random vars and MI is for distributions. Nonetheless I think there can be merit in exploring these
PS: I am the author and just wanted to talk about the familiar basic stuff on this one