We visited iFlytek’s headquarters in Hefei and hosted an AI paneldiscussion with management (GCC). The company shows its ambitionin the AI space and our key findings are: 1) iFlytek enjoys meaningfulshare in the China speech recognition market thanks to the strong base of itsiFLY speech-to-text solution and Siri-like YuDian; 2) the adoption of variousunmanned client support services for telecom operators, financial institutions,and law enforcement agencies; 3) speech recognition devices includingportable and vehicle use (Figs 3-6); and 4) the extension into healthcare anddriving with AI-based cognition and pixel-level image labeling abilities.
Speech recognition is one of the key ways to access AI: iFlytek utilizesexpertise in machine perception (i.e. speech recognition/synthesis), covering22 dialects nationwide with a success rate of 98% in 2017 vs. 60% in 2010,to launch multiple smart devices and services for applications includingtelecom networks (10% of 1H17 revenue), unmanned client support (5%),education (30%), Smart City (5%), etc. For example, iFlytek is introducing anAI-driven voice assistant to replace humans in areas such as client support fortelecom operators, financial institutions and law enforcement agencies,achieving better user experience with 50% less time and 30% more efficiency.
AI + healthcare & driving to be enabled by Deep Learning (DL) extension:iFlytek recently established a remarkable breakthrough in AI-driven healthcareand driving with DL capabilities beyond simple speech recognition (Figs 7-8).
Healthcare – iFlytek’s medical assistant passed the ‘2017 clinician writtenexam’ with a score of 456, well above the qualification requirement of 360,supported by cognition and problem-solving abilities. Driving – the company’sspeech recognition technology has been widely adopted by auto brands suchas Nio, Weltmeister and VW. It has also ventured into image recognition withpromising results in pixel-level semantic labeling tasks.
Read-across to China robots: 90% of service robot makers in China useiFlytek’s speech recognition and synthesis technology, which managementthinks to be a key communication intermediate between machines andhumans rather than touch screens in the future. We estimate the globalservice robot market to reach US$100bn by 2025E with a 32% CAGR in2017-25E on current low base and diversifying applications, driving thesupply chain with more complex functions (Global Service Robots/ AI).
+42% YoY in 4Q17 net profit based on guidance midpoint: Managementguides 4Q17 net profit to be within a range of -17% ~ +100% YoY, with themidpoint landing at +42% YoY. With a commitment to continuous technologyenhancement, the company targets to raise opex for not only R&D but alsotalent recruitment to enlarge its leading position in speech recognition (opex% at 45% in 9M17 vs 42%/39% in 2016/15A). In terms of respective businesssegments, the company expresses strong confidence that the educationsegment will outperform given higher penetration and visibility.