Fine Tuning problem, final accuracy won't go up

[En Liu]

HI,

Firstly, thanks for Yamnet to give us a chance to leverage the transfer learning in audio area. 

We've encounter a bottleneck that by using the example code shared by Tensorflow example site,  our accuracy stuck around 86% and never go up. Ironically, our old 2019 CGRU model can even get better result (96%) with same dataset. 

Is there any in-depth tutorial or materials that we could explore and improve the final accuracy? 

Thanks In Adv!

Brendan