search

awni / speech

A PyTorch Implementation of End-to-End Models for Speech-to-Text
Apache License 2.0
751 stars 176 forks source link

TIMIT PER #57

Closed ankitapasad closed 4 years ago

ankitapasad commented 4 years ago

With the recommended Seq2seq config, I get the Timit PER of 28% on the test set (instead of the reported 18.7%). Is there anyone else with a similar experience and/or know what could be going wrong?

Thank you!

biyoml commented 4 years ago

Same here... My TIMIT PER is 26.1% using the Seq2seq model... The author mentioned that his results are from "an earlier version" of the training set, instead of the 462 speaker training set. (https://github.com/awni/speech/tree/master/examples/timit#results) There is a discussion on this issue: https://github.com/awni/speech/issues/22

ankitapasad commented 4 years ago

Thanks for your reply and the reference!