Help for input and output shape of my data set for Regression

can anyone help me regarding my FYP project.I am building a text to speech model using my own data set . I have text lines as input of shape(868,70) where 70 is the max words in each record. And Audio Features as output of shape (868,82688) each record has Mel-Spectrogram features .

I want to know which activation and loss function I use in my layers to get only 82688 unique values in output. for example [1.6796985,0.4172823,0.0011608523,…,0.000051741976]

I know it is a regression problem. Please help.