At CES 2018, KikaGo, a voice powered platform, has become one of the hottest technologies at the show. It was recognized by the Wall Street Journal as “Best of CES 2018,” and also won the 2018 CES Innovation Awards for Smart Home, Software and Mobile Apps, In-Vehicle Audio/Video, and Tech for a Better World. In the era of AI, KikaGo is much more than your traditional phone cable.
KikaGo is not only a smart cable, but also a “full driving assistant.” In addition to charging the phone, KikaGo helps users communicate completely hands-free while driving, including: taking calls, receiving and sending messages through platforms including SMS, Facebook Messenger, navigation, and playing music.
What’s even more special is that KikaGo can accurately identify the user’s voice inputs in a very noisy driving environment and through voiceprint recognition technology, listening to only the driver’s voice. In a typical driving environment, the distance between the driver and mobile phone is about 50 cm. To interact with your phone voice at this distance, you need a loud voice. KikaGo solves this problem by creatively adding two microphones with opposite direction on the common data charging cables. In order to solve the problem of noise reduction, at least two audio signals are needed. There is also a chip on the data cable for signal conversion and gain. These acoustic signals are then transmitted to the KikaGo application on the phone to continue completing speech recognition, and semantic understanding.
Full voice solution
What are the technical challenges behind KikaGo? The first and most immediate challenge is accent. Similar to dialects in phonetic interaction in China, the English language also faces accent problems. The common solution is to collect larger accent data, and then carry out the training to optimize the model, or use less data, but use more deep neural networks to achieve better results.
Second, there are understandings. Unlike text input, there is a lack of punctuation, emoji, and other information that reflects human emotions in speech interactions, which can easily lead to misunderstandings. In terms of emotional understanding, Kika builds an emotional dataset that covers a variety of emotional expressions, find a connection between the user input voice and the emoji series, and use deep learning and network training. The size of the data set is close to 100 million, mainly from anonymous desensitization data entered by Kika users.
Cars with different spaces, such as big cars, small cars, SUV, etc., may have an impact on the effect of radio. How to achieve the best recognition effect in different spatial distance, and how to reduce the noise in the scene such as fast speed, opening windows, playing music inside the car are all the difficult problems that need to be polished and tested.
This is the first time that Kika has launched a hardware-related product. Given the current market feedback, four CES Innovation Awards, the Wall Street Journal distinction as a Best of CES 2018, a nomination for the Most Innovative App category in the 2018 Global Mobile Awards (GLOMOs) from MWC Barcelona, and feedback from users on social media, KikaGo provides an innovative solution for its users.
For human-computer interactive products represented by input methods, the core challenges are: whole, accuracy and speed, which have not been addressed before the revolution of deep learning. Based on AI technology, in the “whole” aspect through the index, analysis and recommendation in the multimedia content library, the user can feel the word library can be more complete from the input perception level; and on the “accuracy” side, the recognition and understanding has a 60–90 percent improvement. The shortcomings of traditional statistical models can also be set aside on the issue of speed to improve the accuracy of associative words. In addition to the ability improvement of the core products itself, Kika looks to continue to push boundaries and develop products like KikaGo.
Who is Kika Tech?
Kika Tech has dual headquarters in Beijing and Silicon Valley, and establishes R&D centers in India and Taipei, respectively. It launched its first input method product, the Kika Keyboard in 2014, and also got the title of “top developer of Google Play” in 2015. Presently, Kika has entered more than 140 countries overseas and supports 173 languages, as its total number of users in the world reaches 400M, with 60 MAU. In other words, Kika does something similar to Sogou in the country. This may explain why Kika is able to develop KikaGo with its established user base, data, voice input, and human-computer interaction.