iFlytek launches SparkAudio 1.0 voice AI model

iFlytek has launched its new voice foundation model, SparkAudio 1.0, at the Global 1024 Developer Festival. The model combines machine translation, speech recognition, and semantic understanding, giving it much stronger basic voice skills.

It performs well in real-time speaker separation, environment understanding, audio analysis, and emotion detection. The company showed a demo with a movie clip, where the model accurately picked out voices, backgrounds, speaker gender, and scenes, pushing AI closer to real-world use.

The model is now live in iFlytek Input Method 16.0, its transcription app, office notebook, and translator devices, with free Chinese-English switching and real-time speaker detection. The company also released an AI recorder ring with a 4+1 mic array and Wi-Fi 6. It claims over 96% accuracy and can turn a 2-hour recording into text in just 2 minutes.