A month ago, Microsoft's conversational speech recognition technology achieved a breakthrough in word error rate (WER) as low as 6.3% in the industry standard Switchboard speech recognition benchmark, creating the lowest error rate in the field at that time.
Recently, Microsoft further reduced the word error rate to 5.9%, which was the first time to achieve the same level as a professional stenographer and better than most people.
Microsoft's speech recognition technology can refresh records again and again, thanks in large part to the open source tool CNTK (ComputaTIonal Network Toolkit).
The system was originally developed to study voice applications and later expanded into a Microsoft localized deep learning system. The CNTK toolkit was open sourced on GitHub a year ago, and the current speech recognition including Microsoft's artificial intelligence personal assistants Xiaona and HoloLens is based on CNTK. The biggest difference between CNTK and other open source software is that it can do large-scale, distributed machine learning while ensuring powerful performance.
Today, the toolkit has been updated to include a beta version called the Cognitive Toolkit.
The research on speech recognition can be traced back to a related project funded by DARPA (Defense Advanced Research Projects Agency, US Department of Defense Advanced Research Projects Agency, mainly dedicated to the research, development and application of high-tech defense in the United States) in the 1970s. In the decades that followed, more and more research institutions and large companies joined in.
"This breakthrough is the result of the continuous accumulation of speech recognition technology over the past two decades," said Geoffrey Zweig, a researcher at Microsoft's Language and Dialogue Research Group.
What is the concept of 5.9% word error rate that Microsoft is now achieving? In the industry standard Switchboard speech recognition task test, the human control group (composed of professional stenographers) transcribes conversational speech into text. The lowest recorded error rate is 5.9%, which means the voice of Microsoft's speech recognition system. The recognition ability has been higher than that of the vast majority of people in the world and has been tied to the human profession, creating a new world record.
The milestone breakthrough in speech recognition will have a profound impact on consumers and commercial products, as speech recognition technology can significantly enhance people's everyday computing experience. These products include entertainment devices like XBOX, productivity tools like Microsoft's Cortana, and personal artificial intelligence assistants that enable real-time voice-to-text transcription.
Although the Microsoft team's word error rate is 5.9% equal to that of human professional stenographers, this does not mean that the computer can perfectly recognize every word. If the computer can perfectly identify "blue thin" and "champignon" one day, the speech recognition technology will reach a new height.
Storm Circuit has been producing Multilayer PCBs for over 10years. Over the years, we have seen all types of multilayer constructions from various industries, answered all types of multilayer questions, and solved all types of problems with multilayer PCBs.
Multilayer PCB is a circuit board with two or more layers. The material unlike Double-Sided PCBs with only two conductive layers all multilayer PCBs is that it must have at least three layers of conductive material, and the PCB boards are buried in the center of the material.
Benefits of Multilayer PCBs (compared to single or double-sided PCBs)
Higher assembly density
Smaller size (considerable savings on space)
Increased flexibility
Easier incorporation controlled impedance features.
EMI shielding through careful placement of power and ground layers.
Reduces the need for interconnection wiring harnesses (reduces overall weight
Multilayer PCB,Multilayer Bga Board,Multilayer Smd PCB,Multilayer Circuit Board PCB
Storm Circuit Technology Ltd , https://www.stormpcb.com