Jingzhunxue debuts world’s first open-source, end-to-end and education-focused speech LLM FlowMirror
Hangzhou-based AI startup looks to set new standards in AI-driven educational assistance
HANGZHOU, China, Sept. 23, 2024 /PRNewswire/ — Jingzhunxue, a Hangzhou-based education technology company, has launched “FlowMirror-s(V02),” the world’s first open-source end-to-end large language model (LLM) for speech detection and interaction catering to the education industry, marking the company’s endeavor to promote industry knowledge-sharing and the adoption of AI-powered learning services.
In a departure from traditional Automatic Speech Recognition and Text-to-Speech processes, “FlowMirror-s(V02)” is built from the ground up using a self-supervised Chinese speech codec system, with its weight initialized based on a text LLM.
The result is a model trained end-to-end on speech and dialogue data, allowing for low-latency, end-to-end speech interactions.
Jingzhunxue has also developed the world’s first “Hyper-Realistic AI One-on-One Tutor” using its FlowMirror-s(V02) model. Designed under the “AI Native” principle, this AI tutor closely mimics real teachers, offering personalized, systemized instruction that surpasses traditional AI tools and extends across all scenarios.
“We focus especially on educational applications. Hopefully, the success of this model will not only drive innovation in the Chinese AI community but also advance the education sector,” says Renbin Yang, founder and CEO of Jingzhunxue. “This is our cornerstone step toward intelligent learning assistance.”
By leveraging Alibaba’s LLM Tongyi Qianwen (Qwen), FlowMirror stands as one of China’s most advanced educational models. Powered by Qwen’s trillion-parameter framework and trained using the multi-GPU Bailian platform from AliCloud, FlowMirror offers a range of specialized features.
It incorporates over 2 billion proprietary tokens through a sophisticated data pipeline designed for educational assistance. It achieves multimodal interaction by integrating Alibaba’s visual model, enabling dynamic problem-solving and tutoring support.
The model can detect over 40 emotional and physical states through voice and visual cues. Besides, thanks to 160,000 hours of educational speech training, it aligns its speech patterns with those of human teachers.
Additionally, FlowMirror offers personalized teaching styles inspired by renowned educators and utilizes virtual teacher technology to generate exclusive, high-definition, real-time interactions from just one hour of video data. Moreover, its knowledge graph draws on millions of data sets extracted from an extensive question bank and learning data pool to master various educational materials.
FlowMirror-s v0.1 and v0.2 were pre-trained on 20,000 and 50,000 hours of speech data, respectively, validating the model’s end-to-end speech capabilities and scalability.
The latest version, “FlowMirror-s(v0.2),” facilitates seamless speech input-to-output interactions, making it better-suited to educational scenarios with natural, real teacher-like dialogues. Jingzhunxue plans to showcase its technology’s practical applications soon.
In May, Jingzhunxue secured nearly RMB 200 million ($28 million) in a funding round backed solely by Alibaba. The proceeds will be used to develop and promote its AI Native educational devices, such as the Bong series e-learning tablets, now available on e-commerce platforms like Tmall and JD.com.
FlowMirror-s is accessible via GitHub:
https://github.com/Zhejiang-Jingzhunxue/flow_mirror
About Jingzhunxue
Backed by Alibaba, Jingzhunxue specializes in integrating AI software and hardware solutions tailored to the education industry. The company aims to provide AI-driven learning experiences that rival or surpass human instruction, while also making these technologies affordable for all.