Thank you to everyone who developed this model.
#7
by abbas78 - opened
Thank you for sharing this impressive model. I’ve tested it extensively in Turkish, and it works flawlessly. So far, this model has delivered the best performance among open-source options. It follows English instructions very well. I’ve narrated YouTube tutorial videos and dubbed movie scenes. The results are excellent.
Dubbing is a much harder test than narration, so that's a useful data point. Narration can run long without anyone noticing, but a dubbed line has to fit the shot, and most TTS drifts on duration once the sentence gets long.
Curious how it handled Turkish suffix chains. Long agglutinative words are usually where stress placement falls apart.