1. Mentex: create a downsampled dataset. 3. Create an upsampling model that regenerates upsampled audio files as in the picture "Denoising with AEs" 4. Test the upsampling model on a real 8 kHz call recording.