Custom Neural Voice: Updated the training pipeline to UniTTSv3 with which the model quality is improved while training time is reduced by 50% for acoustic models.Reduced pronunciation errors in Hebrew by 20%.In this article Text-to-speech 2021-July release
