Fetching the paper…

Truly unsupervised acoustic word embeddings using weak top-down constraints in encoder-decoder models · Around