Deepgram has announced the release of five new language models and three new use case speech models, along with improvements to their Hindi and Spanish language models and Conversational AI and Meetings models. These enhancements aim to improve accuracy for transcription applications. The company's transfer learning approach and End-to-End Deep Learning architecture have enabled them to achieve high accuracy on these new speech models. Deepgram has also made its French and German language models free for the first 30 days of use, allowing users to collect more audio data for training and improvement.