Kedas xunfei launches sparkaudio 1.0 voice model for real-time speaker recognition and multilingual support

On October 22, 2026, Kedas Xunfei unveiled its first SparkAudio 1.0 voice model during its global 1024 Developer Festival. The model integrates machine translation, speech recognition, and semantic understanding, significantly enhancing basic voice capabilities. It claims to excel in real-time speaker separation, environmental understanding, audio comprehension, and emotion recognition, advancing AI applications in the physical world. A movie clip analysis demonstrated the model's precision in identifying voices, backgrounds, and speaker gender, as well as its scene judgment accuracy. SparkAudio 1.0 is now available in Xunfei Input Method 16.0, Xunfei Hearing, Xunfei Office Pad, and Xunfei Translation Machine, supporting seamless switching between Chinese and English and real-time speaker recognition. Kedas Xunfei also launched the Xunfei AI Recorder Ring, featuring a 4+1 ring microphone array and Wi-Fi 6 transmission, with an overall recognition accuracy exceeding 96%. Two hours of audio transcription can be completed in just two minutes. The company highlighted the device's efficiency and accuracy, positioning it as a key component in its broader AI ecosystem. This release underscores Kedas Xunfei's commitment to expanding AI capabilities in real-world scenarios through advanced voice technology.