The application of voice recognition to control electrical equipment has become an important development trend, but voice recognition systems usually have to be connected to the Internet. As long as it is off the Internet, it will be useless. This article will introduce two sets of offline voice recognition technologies to get rid of the network limitations of voice recognition technology.
Voice recognition has become the main development trend of human-machine interfaces
The deep learning of voice recognition technology through artificial intelligence technology made a breakthrough in 2013 and opened the third wave of development growth. From the perspective of human-machine interaction changes, voice interaction mode will occupy an important position for a long time after this technical breakthrough. Whether it is consumer electronics, communication electronics, automotive electronics, medical electronics, industrial electronics and so on, as long as there are buttons, keyboards, remote controls, touch screens, and/or gesture control areas, there will be voice needs in theory.
Because voice recognition technology relies on big data to perform voice recognition and semantic analysis and apply the corresponding response, leading to a very large database, the current mainstream products like smart speakers must be connected to the network to carry out voice recognition control. In case of a network disconnection, the whole system cannot be used and the network operation is also limited by the network speed and other factors, which often have time differences and slower reactions. Besides, the system must be connected to the network, so there are security risks of invasion. Many electrical appliances in fact only need to understand several fixed voice commands issued by the user, and do not need to be like an omniscient know-all. Therefore, these online voice recognition systems seem to be overqualified.
In order to solve these problems, offline voice recognition technology will be an alternative. It uses a built-in voice recognition system in the chip. The whole system doesn’t rely on the network, which can recognize the voice commands of the user and make the corresponding action. Below, we will introduce two sets of offline voice recognition systems and the development trend of the related applications.
Toshiba electronic device and storage corporation has been committed to the development of artificial intelligence technology for many years. In 1967, the postal code mail classification system was developed, followed by continuous research on the recognition system of text, voice and image. The company has made unremitting efforts in the field of artificial intelligence. Toshiba introduced the TZ2100 embedded system, which can directly embed the voice recognition system into electrical appliances. Users can directly control household appliances through their voice, so that the household appliances respond to the voice commands more quickly and the system is relatively simple, without network interconnection or the risk of invasion.
Toshiba's TZ2100 embedded system currently supports Japanese, Chinese and American English voice recognition. It uses ARM CortexA9 @300MHz processor computing power, supports voice and displays a human-machine interface (HMI), and integrates all the key functions required by 2D graphics display, built-in audio Codec and 1MB SRAM to save the procurement cost of audio Codec and DRAM for customers, and adopts 2 layers of printed circuit boards to save BOM costs for customers.
Toshiba has provided the evaluation reference board and software development kit (SDK) for customer evaluation at present. Customers only need to integrate the TZ2100 with their existing electrical appliances to make the electrical appliances have the function of voice control. Toshiba plans to launch the voice trigger module in the future to simplify the whole system and reduce more costs.
Founded in 2015, Chipintelli Technology Co., Ltd. located in Chengdu, China, is actively engaged in the development of voice recognition integrated circuits and voice recognition application schemes. Chipintelli is positioned in the basic chip layer service market and has great advantages in the field of chip design and voice recognition.
In 2016, Chipintelli launched the CI1006 chip, which integrates an MCU and UART, I2C, SPI, PWM, infrared and other peripheral control interfaces, and supports the integration of local voice recognition and device control. It has high real time performance, and its recognition accuracy is comparable to the cloud. The voice processing algorithm chip can reduce the system power consumption greatly. The CI1006 chip is small in size, DDR, NAND and other devices are not required in the periphery, and the cost of BOM is low.
The CI1006 chip is based on the ASIC architecture, containing the neural network processing hardware unit, which can fully support DNN (Deep Neural Networks) computing architecture and carry out high-performance data parallel computing. Its MCU kernel and various commonly used control interfaces can be used to design various intelligent voice product programs and to realize the single chip local off-line large vocabulary recognition and other product solutions. It is widely used in intelligent appliances, robots, intelligent toys and other product fields. The CI1006 chip is an off-line recognition system, suitable for most electronic products without APP application requirements, but it can also be combined with online recognition and APP schemes to carry out complex multi-functional and multi-task applications, which is suitable for projects with low reliability, power consumption, and cost requirements.
In fact, all kinds of voice recognition schemes have their advantages, whether it is off-line recognition or online recognition, and whether the system needs to be combined with APP operation or not, it should take everything into account based on the actual application requirements and the cost requirements. In any case, the voice recognition control will become the most important development trend of the human-machine interface. We will develop the related products as early as possible to avoid their absence in the market development trend.