Microsoft Releases Open Source Toolkit That Understands Words Just Like Humans, Adds C++ & Python Support

Microsoft flaunted their new speech recognition system that can recognize words in a conversation to the same extent as a human would do. It was able to achieve a low word error rate (WER) of 5.9 percent. The system is built using the Microsoft Cognitive Toolkit, previously known as CNTK toolkit, an in-house project by Microsoft researchers for their personal use. They created it for speech and image recognition but the advantages of the toolkit were felt and it was taken beyond image and speech recognition.

The cognitive toolkit helps researchers with neural networks leading to the creation of their own machine learning systems which can run on computers with traditional CPUs and GPUs. Data sets of variety in size can be processed by the Microsoft Cognitive toolkit on a single machine or on a series of computers in a data center. The toolkit enables improved performance on Pascal architecture-based GPUs in the Nvidia DGX-1.

The updated toolkit can be used as a library with C++ and Python APIs. It also enables reinforcement learning research for machine learning systems which involves training them to do a particular task by trial and error method. This would allow these AI agents to take complex decisions.

A Germany-based company Liebherr has tried to use the Microsoft Cognitive Toolkit by installing cameras in their refrigerators. The cameras are able to detect what food is present inside and an inventory list can be created automatically. The future prospects indicate shopping and meal planning based on the monitoring of food items.

Microsoft Cognitive toolkit has been used by the Bing team to understand the context of the search queries. For instance, when a user types “How to make an apple pie?”, Bing should be able to understand that the search is made for the recipe even though the word “recipe” isn’t included in the search query.

Microsoft’s latest open source toolkit gives a competition to existing developments like Google’s TensorFlow. The toolkit allows a system to be scalable and reduces the training time. The Microsoft Cognitive toolkit is available on GitHub as beta.

Source: Microsoft

If you have something to add, tell us in the comments below.

Also Read: Future of Machine Learning: Business Implications and Big Data