It's a slightly better model for TTS. With extra training focusing on reading the script exactly as written. e.g. the audio-preview model when given instructio

It's a slightly better model for TTS. With extra training focusing on reading the script exactly as written. e.g. the audio-preview model when given instruction to speak "What is the capital of Italy" would often speak "Rome". This model should be much better in that regard = No plans to have localized voice models, but we do want to bring expand the menu of voices with voices that are best at different accents