I have a question about the trigger word detection assignment
why is GRU used instead of LSTM, what makes the GRU more appropriate for this problem?
Also, is it possible to use both GRU and LSTM in the same model? can this enhance the accuracy?
Lastly, based on what GRU units should be used, why the dropout was chosen as 0.8?
The answer is one word: experimentation.
Yes, we can use GRU and LSTM in the same model. You may try all this and update us with what you get.
Best,
Saif.
2 Likes
I tried to use LSTM instead of GRU and obtained less accuracy, I am not able to get understand why? I thought that LSTM will cause an increment in the accuracy of the model.
Can you please explain the reason for that? what is the difference that makes the GRU better for this problem but not LSTM?
1 Like
Well, I cannot give you the exact reason why LSTM didn’t fit, but I guess it overfitted the data. Do we need any memory to remember just one word: trigger word? Furthermore, different hyperparameters lead to different results.
1 Like