Communicating with machines and having them understand what you are saying has turned the once-dream of voice recognition technology into reality. Voice recognition acts like a machine’s auditory system, allowing it to convert voice signals into corresponding text or commands through recognition and understanding. Since the first experimental system capable of recognizing 10 English digits was developed by Bell Labs in 1952, voice recognition technology has undergone groundbreaking development, no longer limited to isolated word recognition. Now, it can achieve continuous multilingual real-time recognition. In real life, voice recognition products have become very common, such as the Siri application on iPhones, Google Now developed on the Android system, Baidu Voice, and Microsoft’s cheeky Cortana voice recognition software. However, most of the popular voice recognition solutions on the market are software-based; below we will introduce a language recognition board based on FPGA implementation.

With just one month left in the Indiegogo project fundraising, MATRIX Voice’s open-source voice platform has exceeded its expected goal of $5000, reaching an impressive 289%. MATRIX Voice is the third crowdfunding project initiated by MATRIX in Miami, Florida. In fact, the MATRIX Voice platform is a 3.14-inch circular circuit board, which is not only compatible with the latest language libraries but also capable of real-time continuous language recognition. It has received recognition from several cloud-based language services, including Microsoft’s recognition server, Amazon’s Alexa voice service, Google’s Speech API, as well as Wit.ai and Houndify. Structurally, MAXTRIX Voice is based on Xilinx’s Spartan-6 LX4 FPGA implementation. Its design allows it to be plugged directly into a low-power Raspberry Pi single-board computer or used as an independent board.

The following two images show the top and bottom views of the MATRIX Voice:

Figure 1 MATRIX Voice top view

Figure 2 MATRIX Voice bottom view

From the images above, it can be seen that the back of the MATRIX Voice board has seven MEMS microphones, seven REG multicolor LEDs, and the Spartan-6 FPGA chip is also on the front panel. On the back of the MATRIX Voice, there is a 64Mbit SDRAM and a set of I/O interfaces compatible with the Raspberry Pi single-board computer. As this is the latest version in the MATRIX implementation series of development boards, there is already a relatively complex layered software stack support for the MATRIX Voice board, including a Hardware Abstraction Layer (HAL) for transparent FPGA code and a C++ function library, as well as an intermediate layer between the video stream interface and the vision library (mainly for the Raspberry Pi camera). In addition, the MATRIX operating system and high-level API interface are provided for the top layer of MATRIX Voice. When developing based on MATRIX Voice, users can choose their preferred programming language according to their habits, mainly because the MATRIX Voice software library supports a variety of languages, including mainstream C++, Python, JavaScript, and commonly used hardware languages Verilog and VHDL. The following image shows the software development layers of the MATRIX Voice platform:

Figure 3 Software layer view

Now, crowdfunding projects in the field of science and technology are becoming increasingly common. Developing cutting-edge applications based on FPGA with greater functionality and flexibility is a very good idea, as it not only provides products with richer interfaces but also allows for the integration of targeted accelerator modules, making the final product not only high-performance but also cost-effective. Among the FPGA chips developed by Xilinx, there are many that can provide both software programming and hardware programming, so choosing Xilinx will make your products more reliable and powerful.