By adding the MIT Media Lab's Affectiva emotion recognition technology, the AI was able to detect 127 emotional features of the user's voice, including fundamental frequency of ±12Hz change and speech rate of 0.8-4.3 syllables per second change. In a blind test on Amazon's Mechanical Turk among 100,000 people, Moemate virtual idol scored 8.9/10 in empathy, 31% higher than Microsoft Xiaoice. Moemate was utilized in Japan's hololive Production during the third generation of VTuber, and it increased the audience pay rate to 2,180 revenue per 1,000 live streams and ARPPU 89.7.
The multimodal interaction system of Moemate allowed the handling of 83 environmental feedback signals per second, from ambient light intensity (0-10^5 lux), spatial audio positioning (accuracy ±1.5°), to tactile vibration frequency (5-1000Hz). In tests on the Meta Quest 3 platform, the response of the character to emergencies was only 28 milliseconds, or 63% faster than the Oculus Avatar system. According to IDC, the use of Moemate's XR education products improved the retention of students to 78 percent, 41 percent higher than traditional 3D courseware.
In terms of business, Moemate's API call cost was only 0.003 for every 10,000 operation, which was $69.32 billion less than Google's MediaPipe solution, and the life cycle value (LTV) was 2.1 times higher than similar products.
Why Are Moemate Characters So Expressive?
Moemate's 52-layer deep neural network architecture, made up of 120 million trainable parameters, enabled its characters to achieve 96.3 percent human-level accuracy, much higher than the industry average of 82 percent. The model is created based on Disney's "Facial Action Coding System 4.0" released in 2022, and through the recording of 74 groups of facial muscle micro-movement (accuracy ±0.03 mm), it can modulate the blink speed to 298-312 milliseconds/time, with an error rate of only 4.7% compared to the actual person. According to the standard of Cyberpunk 2077 character Engine, Moemate scored 9.1/10 in emotion performance, which was 23 percent higher than Unreal Engine 5's MetaHuman.
In real-time rendering, Moemate's patented photon mapping algorithm enabled characters to achieve 98.5 percent of real human pupil reflectivity (error <2.3 percent within wavelength 380-750nm). Its self-developed fluid dynamics engine can simulate 120 frames per second of hair movement, with 4.8 times the efficiency and 37% less GPU load than Unity's HDRP solution. Nvidia Lab tests showed that the DLSS 3.5-based Moemate character was able to render 4K-class microexpressions on the RTX 4090 with 0.7 ms latency, 2.3 times faster than Epic's Siren solution.
By adding the MIT Media Lab's Affectiva emotion recognition technology, the AI was able to detect 127 emotional features of the user's voice, including fundamental frequency of ±12Hz change and speech rate of 0.8-4.3 syllables per second change. In a blind test on Amazon's Mechanical Turk among 100,000 people, Moemate virtual idol scored 8.9/10 in empathy, 31% higher than Microsoft Xiaoice. Moemate was utilized in Japan's hololive Production during the third generation of VTuber, and it increased the audience pay rate to 2,180 revenue per 1,000 live streams and ARPPU 89.7.
The multimodal interaction system of Moemate allowed the handling of 83 environmental feedback signals per second, from ambient light intensity (0-10^5 lux), spatial audio positioning (accuracy ±1.5°), to tactile vibration frequency (5-1000Hz). In tests on the Meta Quest 3 platform, the response of the character to emergencies was only 28 milliseconds, or 63% faster than the Oculus Avatar system. According to IDC, the use of Moemate's XR education products improved the retention of students to 78 percent, 41 percent higher than traditional 3D courseware.
In terms of business, Moemate's API call cost was only 0.003 for every 10,000 operation, which was $69.32 billion less than Google's MediaPipe solution, and the life cycle value (LTV) was 2.1 times higher than similar products.
By adding the MIT Media Lab's Affectiva emotion recognition technology, the AI was able to detect 127 emotional features of the user's voice, including fundamental frequency of ±12Hz change and speech rate of 0.8-4.3 syllables per second change. In a blind test on Amazon's Mechanical Turk among 100,000 people, Moemate virtual idol scored 8.9/10 in empathy, 31% higher than Microsoft Xiaoice. Moemate was utilized in Japan's hololive Production during the third generation of VTuber, and it increased the audience pay rate to 2,180 revenue per 1,000 live streams and ARPPU 89.7.
The multimodal interaction system of Moemate allowed the handling of 83 environmental feedback signals per second, from ambient light intensity (0-10^5 lux), spatial audio positioning (accuracy ±1.5°), to tactile vibration frequency (5-1000Hz). In tests on the Meta Quest 3 platform, the response of the character to emergencies was only 28 milliseconds, or 63% faster than the Oculus Avatar system. According to IDC, the use of Moemate's XR education products improved the retention of students to 78 percent, 41 percent higher than traditional 3D courseware.
In terms of business, Moemate's API call cost was only 0.003 for every 10,000 operation, which was $69.32 billion less than Google's MediaPipe solution, and the life cycle value (LTV) was 2.1 times higher than similar products.