This has two parts.
1.Server API: The server should be able to receive a list of Ids and return the low level data for the same.
2.Client Side: The client should make the request and save these files in a temporary location, which will be used for evaluation.

We are actually unsure of the best thing to do here, but for now what we will be doing is that the client will be making requests of about 100 songs in a single request, receiving the server response and the saving them to temporary files which will be further used in the dataset training. We also thought of transferring the full dataset as a compressed file, but that would keep the connection opened for too long. So if anybody has any better solutions please discuss them in the comments.