This datasheet explains Deepgram’s four-tier speech model framework: Beginner, Intermediate, Advanced, and Expert. Each tier aligns with different accuracy goals, training complexity, and word error rate (WER) reduction targets. Beginner models support general use cases, while Advanced and Expert tiers leverage AutoML and supervised AI scientist involvement for complex audio environments, new languages, and regulated industries. Organizations can reduce WER by up to 8% or more depending on training depth. The sheet outlines time-to-create estimates and guidance for aligning model selection with business outcomes.

Join for free to read