search

soobinseo / Transformer-TTS

A Pytorch Implementation of "Neural Speech Synthesis with Transformer Network"
MIT License
659 stars 140 forks source link

Clarification on experiments #33

Open bkumardevan07 opened 3 years ago

bkumardevan07 commented 3 years ago

Hi, Can you clearify what you exactly mean when you say .. "It was very important to concatenate the input and context vectors in the Attention mechanism."

Also could you also specify your stopping criteria as you mentioned you did not use stop loss.

soobinseo commented 3 years ago
  1. "It was very important to concatenate the input and context vectors in the Attention mechanism."

When the experiment was conducted, it was found that the attention plot was not properly formed when the layer was passed without the above conditions, and the quality of the result was not good.

  1. when I use stop loss, the training was not going well. (I should find why it wasn't work)