At the main forum of the 2026 iFlytek Global 1024 Developer Festival, Chairman Liu Qingfeng unveiled the SparkAudio 1.0 voice foundation model, multimodal interaction technology, and Spark embodied multimodal interaction technology along with related products, announcing a continued progression along the path of "voice interaction to multimodal interaction to embodied multimodal interaction," aiming to lead the next generation of human-machine interaction through integrated software and hardware innovation.
In the field of voice interaction, high-noise recognition has long been an industry challenge, and complex audio perception and multi-speaker voice analysis built on top of it are even more advanced capabilities. To address these difficulties, Iflytek Co.,Ltd. (002230) released the SparkAudio 1.0 voice foundation model, delivering comprehensive upgrades in speech recognition, simultaneous interpretation and interaction, as well as audio understanding, emotion recognition, real-time speaker separation and environmental sound comprehension. It is reported that SparkAudio 1.0 possesses capabilities similar to a "sound detective" in crime dramas—even in a noisy street, it can distinguish horns, street vendors' calls, and the sound of wheels rubbing against the ground, separate overlapping voices, and infer the surrounding environment from sound clues. It not only "hears what is said" but also "understands what is happening." Related speech recognition tests show an overall performance improvement of 20%, a 15% reduction in response time, and a 60% cost reduction, making voice services more accurate, more timely, and easier to scale.
Currently, SparkAudio 1.0 has already been deployed across iFlytek products, and users can experience it in iFlytek Input Method 16.0, iFlytek Hearing, iFlytek Smart Office Notebook, and other products. The event also saw the launch of the iFlytek AI Recording Ring, which can magnetically attach to the back of a phone like a phone stand, enabling recording at any time. Two hours of recording can be transcribed in just two minutes, with comprehensive recognition accuracy exceeding 96%.
In addition, iFlytek released multimodal interaction technology, further integrating multimodal far-field recognition, multi-person interaction, audiovisual fusion understanding, personalized memory, and emotional semantics to allow AI to understand richer contextual information. The iFlytek AI Glasses are a typical application product of this technology. At the event, Lin Huijie, General Manager of iFlytek's Wearable Business Unit, demonstrated meeting notes from Liu Qingfeng's speech that had just been recorded using iFlytek AI Glasses, with clear content structure and rich text-image integration.
Moving from intelligent terminals to mobile robots, interaction must also adapt to spatial changes, multi-person participation, and action execution. To address these challenges, Iflytek Co.,Ltd. (002230) released Spark embodied multimodal interaction technology, achieving breakthroughs in complex spatial dynamic noise reduction, multi-target human interaction intention recognition, unified planning and execution of speech and actions, and vision-triggered proactive interaction. This enables robots to both communicate with people and collaboratively complete tasks in dynamic environments, providing key technical support for their entry into home and social service scenarios.
Iflytek Co.,Ltd. (002230) also launched the Robot Super Brain Platform V3.0, integrating the Spark Super Brain Board, embodied intelligence development kit, multimodal interaction modules, and embodied multimodal interaction Harness. Through device-cloud collaboration and front-back-end collaboration, it enables robots to chat continuously on the front end while handling tasks continuously on the back end, and supports modular and configurable industry customization.
Comments