Create basic verification by using already computed embeddings

[MatthewGree]

Develop basic verification based on embeddings.

Given data:

• the sound embeddings of all samples from the training dataset
• mapping of the mentioned embeddings to user_id
• the sound embedding of sample that needs verification
• the user_id which came with the mentioned sample

The verification should produce True if the user verification succeeded and False otherwise.

It should base on simple methods, like using cosine similarity and threshold.