SeaVoice currently supports 12 languages. The English and Mandarin (Taiwan) models are our own in-house models trained from scratch; they are highly accurate and reliable. However, we’re always working on creating new language models and improving our existing ones. We’d love to hear which languages you’re most eager to use! All other languages are supported using a multilingual open source model as the base. The performance wasn’t great out of the box, so we integrated it into our own STT pipeline and tuned the model to improve the performance. One thing you may notice with the open source model is “hallucination”. This can manifest in a couple different ways, such as: inserting words/phrases that weren’t said, transcribing in the wrong language, and/or translating the spoken language to a different language.