iFLYTEK In-Car Voice System

Definition and Technical Positioning

iFLYTEK is a leading player in China's in-vehicle voice interaction market and a renowned listed intelligent speech and artificial intelligence firm in the Asia-Pacific region. Since embarking on its automotive intelligence journey around 2015, the company has deeply cultivated core technologies including intelligent voice, natural language processing, and cognitive intelligence, delivering full-stack solutions ranging from intelligent interaction to smart cockpits for the automotive industry. According to statistics from Gasgoo Automotive Research Institute, iFLYTEK's in-vehicle voice interaction products ranked first in China in 2025 with a 41.9% market share, achieving an installation volume of 8.188 million units.

Technical System: On-Device Large Model and Hybrid Architecture

At the 2026 Beijing Auto Show, iFLYTEK unveiled SparkAuto-EMM, an on-device multimodal large model for its Spark Cockpit, offering models spanning 0.5B to 7B parameters to cater to varying platform computing powers and differentiated vehicle requirements. This edge-deployment strategy is designed to overcome four major bottlenecks of pure cloud solutions: high API computing costs, unbearable network latency, functional failure in areas like basement car parks, and data privacy compliance concerns from constant cloud uploading. Through a hybrid edge-cloud architecture, Spark Cockpit maintains core functionality offline while keeping user data processed locally within the vehicle. According to iFLYTEK President Wu Xiaoru, the performance of on-device large models in dedicated automotive scenarios is now close to that of cloud-based models, while offering significantly faster response speeds. iFLYTEK is also set to introduce an even larger 30B parameter model soon, further pushing the boundaries of on-device intelligence.

"Gets Things Done, Truly Understands You": Spark Cockpit 3.0

The core capabilities of Spark Cockpit 3.0 can be encapsulated into three pillars: "understands accurately, remembers well, and gets things done".

On the intent understanding front, SparkAuto-EMM supports over 1,500 cockpit atomic capabilities across various scenarios, boasting an intent recognition response time of just 150 milliseconds and an accuracy rate exceeding 90% for ambiguous requests. Drivers do not need to rely on rigid preset voice commands; natural, vague phrases like "why is it so stuffy in here" can be interpreted accurately as the system factors in contextual cues such as weather conditions and occupant status.

In terms of memory capability, the system integrates a 54-dimensional user memory model with 2,808 memory anchor points to recall preferences of drivers and their family members. For instance, when planning complex multi-stop itineraries like "pick up my wife from her office, head to the nearest supermarket, then go to the mall for dinner and a movie", the cockpit can automatically chart the optimal route.

For execution capability, Spark Cockpit 3.0 introduces the MasterAgent intelligent hub, which breaks down complex requests and orchestrates over 1,500 vehicle functions alongside specialised scenario Agents. A driver can simply say, "I have an 11:00 AM meeting at the office, lunch with a German client at noon—find a reputable German restaurant nearby—and I need to head to the airport at 3:00 PM, but charge the car first," and the system will process the entire sequence seamlessly without requiring step-by-step prompts.

Proactive Services and Multimodal Sensing

Another key upgrade in Spark Cockpit 3.0 is the shift from "passive response" to "proactive service". Leveraging multimodal sensing technologies such as in-cabin Driver Monitoring Systems (DMS), Occupant Monitoring Systems (OMS), and vital signs detection, the cockpit monitors and anticipates occupant needs in real time. Typical scenarios include automatically lowering music volume and adjusting air conditioning airflow when a child falls asleep in the rear seats, alerting occupants if a smartphone slips between seats, or warning if a pet is left behind. By relying on real-time in-cabin environmental perception rather than direct user inputs, the AI cockpit transforms from a mere execution interface into an intelligent travel companion.

Intelligent Sound Field (iFLYSOUND)

In the acoustic arena, iFLYTEK introduced a master-tape-grade in-car audio system powered by Gallium Nitride (GaN) technology in 2026, delivering up to 300W per channel. It is the first in-car hardware certified under QQ Music's "Premium Sound Quality" standard. Integrating over 30 proprietary in-car AI sound algorithms, it supports zoned sound fields, mic-free in-car karaoke, and dynamic accompanying DJ features, with over 1.2 million units installed to date. Models such as WEY V9X, Dongfeng Nissan NX8, ZEEKR 007GT, and Chery EXEED EX7 come equipped with iFLYSOUND technology.

Global Footprint

To support the global expansion of Chinese automakers, iFLYTEK's overseas Spark large model supports 32 languages across 60 countries and territories, offering seamless multilingual switching. Eight of China's top 10 automotive exporters have established deep partnerships with iFLYTEK, deploying the technology across more than 100 export models. To cater to varying overseas market demands in network infrastructure, data privacy regulations, and digital ecosystems, iFLYTEK offers four dedicated tiers—Lite, Pro, Ultra, and Omni—spanning fully offline to hybrid edge-cloud configurations. On the ecosystem front, iFLYTEK collaborates with carmakers to standardise integration with local overseas third-party service providers (such as navigation, streaming music, and lifestyle platforms), minimising redundant development and speeding up worldwide rollouts.

Technical Limitations and Industry Challenges

While iFLYTEK's market leadership in automotive voice solutions is well-established, running on-device large models demands substantial in-car chipset computing power. For entry-level variants with limited processing headroom, on-device intent recognition and response latency may be compromised, necessitating trade-offs with cloud-based or lightweight architectures. Furthermore, while edge AI ensures privacy and offline availability, executing complex tasks requiring real-time external services (such as restaurant reservations or cross-platform price comparisons) still relies on stable internet connectivity and seamless third-party API integration.