Deepgram has expanded its Nova-3 speech-to-text model to support 11 additional languages across Eastern Europe, South Asia, East Asia, and Southeast Asia, addressing challenges that traditional models face with tonal languages, complex word structures, and multiple writing systems. This update allows Nova-3 to adapt to diverse linguistic structures, such as the syllable timing of Japanese, the vowel harmony of Hungarian, and the tonal contours of Vietnamese, without the need for custom pipelines. The model's Keyterm Prompting feature enables developers to control product names and technical vocabulary across these languages, enhancing accuracy in both batch and streaming modes, with significant Word Error Rate reductions, particularly in languages with complex morphology or non-Latin scripts. This expansion underscores Nova-3’s capability to provide scalable, enterprise-grade voice AI solutions globally, offering improved recognition and lower latency in multilingual workflows while promising further growth into new regions and languages.