Data:
Speech enhancment dataset (noise, voice): 3000h Latvian transcripts: 350h (HQ 90h) Latvian text-corpus: 50m