A family of Riva production (NVAIE) speech models that achieve state-of-the-art results on speech transcription, translation, and synthesis tasks.