On 26 October, Pandaily reported that iFLYTEK had introduced SparkAudio 1.0 and Robot Super Brain Platform 3.0 at its 2026 Global 1024 Developer Festival, combining speech tasks in one model and packaging hardware and software for robots.
SparkAudio combines speech recognition, translation and language understanding. A 0.65-billion-parameter audio encoder feeds a 30B-A3B mixture-of-experts language model, which can route work through specialised parts. iFLYTEK says it trained the system on 13 million hours of audio and large text corpora using Huawei Ascend 910B and Hygon BW1000 chips. The company claims recognition across 99 languages and 202 dialects and says the model can separate overlapping voices. Its technical report says instruction following and complex audio reasoning still need work.
Transcribing a conversation with overlapping voices could draw on that claimed separation before turning speech into text.
Robot Super Brain Platform 3.0 includes a board, a multimodal interaction module and an embodied AI development kit. Its software coordinates work between device and cloud, and between interactions and background tasks, so a robot can continue talking while carrying out other work. At the festival, iFLYTEK demonstrated the platform in an unstaffed, 24-hour robot-run store.
.jpg){kind=link}