Xiaomi announced the launch of a new open-source voice model named MiDashengLM-7B, designed to enhance user experiences in smart cars and home devices. This new model represents a significant advancement in AI tools, expanding beyond text processing to include instant and precise voice interaction, opening wider applications for daily life. MiDashengLM-7B is based on Xiaomi’s core voice model and integrated with Alibaba’s open-source Qwen2.5-Omni-7B model, enhancing processing power and use case diversity. According to XiaomiTime, the model excelled on 22 general benchmarks, outperforming competitors in response speed and processing efficiency. The first token response time is 25% faster than the average of similar AI solutions. MiDashengLM-7B can handle 20 times more concurrent operations than traditional models without additional memory, making it ideal for environments requiring immediate and efficient performance.

Xiaomi trained the new voice model on publicly available data, enhancing transparency and supporting developers in creating easily integrable voice AI tools for various smart systems.