Gradium Launches stt-translate and s2s-translate, Real-Time Speech Translation Models Beating gpt-realtime-translate on Accuracy and Latency

Gradium released two real-time speech translation models that convert speech in one language to text or speech in another language, supporting five languages across twenty language pair combinations. The models use a streamlined two-step pipeline instead of the standard three-step approach, combining transcription and translation into a single pass to reduce processing time. According to Gradium's benchmarks on conversational speech, these models achieve better accuracy than competing systems while maintaining comparable latency, and they include voice control features such as voice selection and cloning that some competing products lack. The models stream results live in a browser over a single connection, eliminating the need to separately integrate multiple translation components.