I try to fine tune SpeechT5-base on clean-train-100 dataset using Transformer library
The problem I am encountering now is that my verification set is too good when it is other-test (wer=2.76), which makes me suspect that there may be problems with the model or method I use, but the effect of clean-test is the same as The papers are similar.
Here is the model I fine-tuned
Can you help me see if there is a serious problem with the experimental part?
Thanks♪(・ω・)ノ
I try to fine tune SpeechT5-base on clean-train-100 dataset using Transformer library The problem I am encountering now is that my verification set is too good when it is other-test (wer=2.76), which makes me suspect that there may be problems with the model or method I use, but the effect of clean-test is the same as The papers are similar.
Here is the model I fine-tuned Can you help me see if there is a serious problem with the experimental part? Thanks♪(・ω・)ノ