Microsoft unveiled a speech recognition breakthrough: a transcription system that matches humans with a word error rate of 5.9% for standard speech. It’s built on an open-source toolkit that Microsoft previously developed. However, there is a major new update to the tool, now called Cognitive Toolkit, that was just released.
Formerly titled the Computational Network Toolkit (CNTK), the MIT-licensed, GitHub-hosted project provides researchers with the necessary foundations, including neural networks, required to develop their own machine learning systems. The applications can run on both CPUs and GPUs, and the toolkit offers support for compute clusters.
CNTK was originally built for speech applications but has since been broadened to provide a wide range of machine learning cases. Bing’s team uses it to draw conclusions about search terms. For example, a search for “How do you make an apple pie?” is a hunt for recipes even if it doesn’t include the word “recipe.” The new version of the toolkit adds features, including support for Python scripting and new algorithms to expand its reach to more diverse applications.
Machine learning applications have become increasingly widespread, with systems such as Cortana and Skype Translator, both of which are dependent on such artificial intelligence techniques. While the computational resources needed to make this type of toolkit are significant, Microsoft released the open source in an effort to “democratize AI” and transform these systems from research projects to real-world applications.
Source: Ars Technica
Advertisement
Learn more about Electronic Products Magazine
