给杰弗里·辛顿做AI翻译的科大讯飞，怎么解决智能硬件体验瓶颈问题？

Group 1 - Jeffrey Hinton, known as the "father of deep learning," delivered a keynote speech at WAIC 2025, supported by iFLYTEK's real-time translation services [1] - iFLYTEK has upgraded its speech simultaneous translation model, achieving a response time of as fast as 2 seconds for English translation, meeting high industry standards [1] - The company has developed a streaming speech synthesis technology that allows for real-time translation and speaking, enhancing accuracy in over 80,000 specialized vocabulary across sectors like healthcare, manufacturing, and finance [2] Group 2 - iFLYTEK is focusing on smart hardware, including recording pens, learning machines, and office notebooks, as key investment areas, with AI translation being a widely applied feature [2] - The smart hardware market is currently characterized by high expectations but low actual performance, attributed to a lack of genuine consumer demand solutions and insufficient integration of software and hardware [2] - iFLYTEK's Vice President Zhao Xiang emphasized that improving product experience in AI translation requires a systemic approach that combines software and hardware optimizations [2] Group 3 - iFLYTEK is enhancing its translation devices to perform well without internet connectivity, necessitating higher computational power on the hardware side [3] - The effectiveness of translation models is significantly influenced by audio capture quality, with iFLYTEK aiming to isolate all sounds beyond one meter in noisy environments [3] Group 4 - iFLYTEK is exploring the integration of multimodal capabilities in translation scenarios to improve transcription and translation outcomes through diverse information collection [5] - The company has achieved coverage of over 50 countries and regions with its simultaneous translation services, but faces challenges in recognizing accents, particularly for non-native speakers [5] - Solutions to accent recognition issues include expanding training data to include both standard and non-standard accents and transitioning to an end-to-end system to enhance semantic understanding [5]