AssemblyAI has significantly improved its speech-to-text features, including punctuation and casing restoration. The company's new model is a multi-class classifier that predicts actions such as adding punctuation or changing casing for each word in the transcription. This transformer-based model architecture yields an accuracy of over 92% for punctuation and casing restoration, trained on over 1 billion tokens. The model performs exceptionally well even with industry-specific language. Punctuation and casing are applied by default to all API requests, making it easy for users to utilize this new feature.